xref: /titanic_41/usr/src/uts/common/os/sunpm.c (revision fef1e07ef354c2dcda4dc397c33f5a5532432c7a)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * sunpm.c builds sunpm.o	"power management framework"
30  *	kernel-resident power management code.  Implements power management
31  *	policy
32  *	Assumes: all backwards compat. device components wake up on &
33  *		 the pm_info pointer in dev_info is initially NULL
34  *
35  * PM - (device) Power Management
36  *
37  * Each device may have 0 or more components.  If a device has no components,
38  * then it can't be power managed.  Each component has 2 or more
39  * power states.
40  *
41  * "Backwards Compatible" (bc) devices:
42  * There are two different types of devices from the point of view of this
43  * code.  The original type, left over from the original PM implementation on
44  * the voyager platform are known in this code as "backwards compatible"
45  * devices (PM_ISBC(dip) returns true).
46  * They are recognized by the pm code by the lack of a pm-components property
47  * and a call made by the driver to pm_create_components(9F).
48  * For these devices, component 0 is special, and represents the power state
49  * of the device.  If component 0 is to be set to power level 0 (off), then
50  * the framework must first call into the driver's detach(9E) routine with
51  * DDI_PM_SUSPEND, to get the driver to save the hardware state of the device.
52  * After setting component 0 from 0 to a non-zero power level, a call must be
53  * made into the driver's attach(9E) routine with DDI_PM_RESUME.
54  *
55  * Currently, the only way to get a bc device power managed is via a set of
56  * ioctls (PM_DIRECT_PM, PM_SET_CURRENT_POWER) issued to /dev/pm.
57  *
58  * For non-bc devices, the driver describes the components by exporting a
59  * pm-components(9P) property that tells how many components there are,
60  * tells what each component's power state values are, and provides human
61  * readable strings (currently unused) for each component name and power state.
62  * Devices which export pm-components(9P) are automatically power managed
63  * whenever autopm is enabled (via PM_START_PM ioctl issued by pmconfig(1M)
64  * after parsing power.conf(4)).
65  * For these devices, all components are considered independent of each other,
66  * and it is up to the driver to decide when a transition requires saving or
67  * restoring hardware state.
68  *
69  * Each device component also has a threshold time associated with each power
70  * transition (see power.conf(4)), and a busy/idle state maintained by the
71  * driver calling pm_idle_component(9F) and pm_busy_component(9F).
72  * Components are created idle.
73  *
74  * The PM framework provides several functions:
75  * -implement PM policy as described in power.conf(4)
76  *  Policy is set by pmconfig(1M) issuing pm ioctls based on power.conf(4).
77  *  Policies consist of:
78  *    -set threshold values (defaults if none provided by pmconfig)
79  *    -set dependencies among devices
80  *    -enable/disable autopm
81  *    -turn down idle components based on thresholds (if autopm is enabled)
82  *     (aka scanning)
83  *    -maintain power states based on dependencies among devices
84  *    -upon request, or when the frame buffer powers off, attempt to turn off
85  *     all components that are idle or become idle over the next (10 sec)
86  *     period in an attempt to get down to an EnergyStar compliant state
87  *    -prevent powering off of a device which exported the
88  *     pm-no-involuntary-power-cycles property without active involvement of
89  *     the device's driver (so no removing power when the device driver is
90  *     not attached)
91  * -provide a mechanism for a device driver to request that a device's component
92  *  be brought back to the power level necessary for the use of the device
93  * -allow a process to directly control the power levels of device components
94  *  (via ioctls issued to /dev/pm--see usr/src/uts/common/io/pm.c)
95  * -ensure that the console frame buffer is powered up before being referenced
96  *  via prom_printf() or other prom calls that might generate console output
97  * -maintain implicit dependencies (e.g. parent must be powered up if child is)
98  * -provide "backwards compatible" behavior for devices without pm-components
99  *  property
100  *
101  * Scanning:
102  * Whenever autopm is enabled, the framework attempts to bring each component
103  * of each device to its lowest power based on the threshold of idleness
104  * associated with each transition and the busy/idle state of the component.
105  *
106  * The actual work of this is done by pm_scan_dev(), which cycles through each
107  * component of a device, checking its idleness against its current threshold,
108  * and calling pm_set_power() as appropriate to change the power level.
109  * This function also indicates when it would next be profitable to scan the
110  * device again, and a new scan is scheduled after that time.
111  *
112  * Dependencies:
113  * It is possible to establish a dependency between the power states of two
114  * otherwise unrelated devices.  This is currently done to ensure that the
115  * cdrom is always up whenever the console framebuffer is up, so that the user
116  * can insert a cdrom and see a popup as a result.
117  *
118  * The dependency terminology used in power.conf(4) is not easy to understand,
119  * so we've adopted a different terminology in the implementation.  We write
120  * of a "keeps up" and a "kept up" device.  A relationship can be established
121  * where one device keeps up another.  That means that if the keepsup device
122  * has any component that is at a non-zero power level, all components of the
123  * "kept up" device must be brought to full power.  This relationship is
124  * asynchronous.  When the keeping device is powered up, a request is queued
125  * to a worker thread to bring up the kept device.  The caller does not wait.
126  * Scan will not turn down a kept up device.
127  *
128  * Direct PM:
129  * A device may be directly power managed by a process.  If a device is
130  * directly pm'd, then it will not be scanned, and dependencies will not be
131  * enforced.  * If a directly pm'd device's driver requests a power change (via
132  * pm_raise_power(9F)), then the request is blocked and notification is sent
133  * to the controlling process, which must issue the requested power change for
134  * the driver to proceed.
135  *
136  */
137 
138 #include <sys/types.h>
139 #include <sys/errno.h>
140 #include <sys/callb.h>		/* callback registration during CPR */
141 #include <sys/conf.h>		/* driver flags and functions */
142 #include <sys/open.h>		/* OTYP_CHR definition */
143 #include <sys/stat.h>		/* S_IFCHR definition */
144 #include <sys/pathname.h>	/* name -> dev_info xlation */
145 #include <sys/ddi_impldefs.h>	/* dev_info node fields */
146 #include <sys/kmem.h>		/* memory alloc stuff */
147 #include <sys/debug.h>
148 #include <sys/archsystm.h>
149 #include <sys/pm.h>
150 #include <sys/ddi.h>
151 #include <sys/sunddi.h>
152 #include <sys/sunndi.h>
153 #include <sys/sunpm.h>
154 #include <sys/epm.h>
155 #include <sys/vfs.h>
156 #include <sys/mode.h>
157 #include <sys/mkdev.h>
158 #include <sys/promif.h>
159 #include <sys/consdev.h>
160 #include <sys/esunddi.h>
161 #include <sys/modctl.h>
162 #include <sys/fs/ufs_fs.h>
163 #include <sys/note.h>
164 #include <sys/taskq.h>
165 #include <sys/bootconf.h>
166 #include <sys/reboot.h>
167 #include <sys/spl.h>
168 #include <sys/disp.h>
169 #include <sys/sobject.h>
170 #include <sys/sunmdi.h>
171 
172 
173 /*
174  * PM LOCKING
175  *	The list of locks:
176  * Global pm mutex locks.
177  *
178  * pm_scan_lock:
179  *		It protects the timeout id of the scan thread, and the value
180  *		of autopm_enabled.  This lock is not held concurrently with
181  *		any other PM locks.
182  *
183  * pm_clone_lock:	Protects the clone list and count of poll events
184  *		pending for the pm driver.
185  *		Lock ordering:
186  *			pm_clone_lock -> pm_pscc_interest_rwlock,
187  *			pm_clone_lock -> pm_pscc_direct_rwlock.
188  *
189  * pm_rsvp_lock:
190  *		Used to synchronize the data structures used for processes
191  *		to rendezvous with state change information when doing
192  *		direct PM.
193  *		Lock ordering:
194  *			pm_rsvp_lock -> pm_pscc_interest_rwlock,
195  *			pm_rsvp_lock -> pm_pscc_direct_rwlock,
196  *			pm_rsvp_lock -> pm_clone_lock.
197  *
198  * ppm_lock:	protects the list of registered ppm drivers
199  *		Lock ordering:
200  *			ppm_lock -> ppm driver unit_lock
201  *
202  * pm_compcnt_lock:
203  *		Protects count of components that are not at their lowest
204  *		power level.
205  *		Lock ordering:
206  *			pm_compcnt_lock -> ppm_lock.
207  *
208  * pm_dep_thread_lock:
209  *		Protects work list for pm_dep_thread.  Not taken concurrently
210  *		with any other pm lock.
211  *
212  * pm_remdrv_lock:
213  *		Serializes the operation of removing noinvol data structure
214  *		entries for a branch of the tree when a driver has been
215  *		removed from the system (modctl_rem_major).
216  *		Lock ordering:
217  *			pm_remdrv_lock -> pm_noinvol_rwlock.
218  *
219  * pm_cfb_lock: (High level spin lock)
220  *		Protects the count of how many components of the console
221  *		frame buffer are off (so we know if we have to bring up the
222  *		console as a result of a prom_printf, etc.
223  *		No other locks are taken while holding this lock.
224  *
225  * pm_loan_lock:
226  *		Protects the lock_loan list.  List is used to record that one
227  *		thread has acquired a power lock but has launched another thread
228  *		to complete its processing.  An entry in the list indicates that
229  *		the worker thread can borrow the lock held by the other thread,
230  *		which must block on the completion of the worker.  Use is
231  *		specific to module loading.
232  *		No other locks are taken while holding this lock.
233  *
234  * Global PM rwlocks
235  *
236  * pm_thresh_rwlock:
237  *		Protects the list of thresholds recorded for future use (when
238  *		devices attach).
239  *		Lock ordering:
240  *			pm_thresh_rwlock -> devi_pm_lock
241  *
242  * pm_noinvol_rwlock:
243  *		Protects list of detached nodes that had noinvol registered.
244  *		No other PM locks are taken while holding pm_noinvol_rwlock.
245  *
246  * pm_pscc_direct_rwlock:
247  *		Protects the list that maps devices being directly power
248  *		managed to the processes that manage them.
249  *		Lock ordering:
250  *			pm_pscc_direct_rwlock -> psce_lock
251  *
252  * pm_pscc_interest_rwlock;
253  *		Protects the list that maps state change events to processes
254  *		that want to know about them.
255  *		Lock ordering:
256  *			pm_pscc_interest_rwlock -> psce_lock
257  *
258  * per-dip locks:
259  *
260  * Each node has these per-dip locks, which are only used if the device is
261  * a candidate for power management (e.g. has pm components)
262  *
263  * devi_pm_lock:
264  *		Protects all power management state of the node except for
265  *		power level, which is protected by ndi_devi_enter().
266  *		Encapsulated in macros PM_LOCK_DIP()/PM_UNLOCK_DIP().
267  *		Lock ordering:
268  *			devi_pm_lock -> pm_rsvp_lock,
269  *			devi_pm_lock -> pm_dep_thread_lock,
270  *			devi_pm_lock -> pm_noinvol_rwlock,
271  *			devi_pm_lock -> power lock
272  *
273  * power lock (ndi_devi_enter()):
274  *		Since changing power level is possibly a slow operation (30
275  *		seconds to spin up a disk drive), this is locked separately.
276  *		Since a call into the driver to change the power level of one
277  *		component may result in a call back into the framework to change
278  *		the power level of another, this lock allows re-entrancy by
279  *		the same thread (ndi_devi_enter is used for this because
280  *		the USB framework uses ndi_devi_enter in its power entry point,
281  *		and use of any other lock would produce a deadlock.
282  *
283  * devi_pm_busy_lock:
284  *		This lock protects the integrity of the busy count.  It is
285  *		only taken by pm_busy_component() and pm_idle_component and
286  *		some code that adjust the busy time after the timer gets set
287  *		up or after a CPR operation.  It is per-dip to keep from
288  *		single-threading all the disk drivers on a system.
289  *		It could be per component instead, but most devices have
290  *		only one component.
291  *		No other PM locks are taken while holding this lock.
292  *
293  */
294 
295 static int stdout_is_framebuffer;
296 static kmutex_t	e_pm_power_lock;
297 static kmutex_t pm_loan_lock;
298 kmutex_t	pm_scan_lock;
299 callb_id_t	pm_cpr_cb_id;
300 callb_id_t	pm_panic_cb_id;
301 callb_id_t	pm_halt_cb_id;
302 int		pm_comps_notlowest;	/* no. of comps not at lowest power */
303 int		pm_powering_down;	/* cpr is source of DDI_SUSPEND calls */
304 
305 clock_t pm_min_scan = PM_MIN_SCAN;
306 clock_t pm_id_ticks = 5;	/* ticks to wait before scan during idle-down */
307 
308 static int pm_busop_set_power(dev_info_t *,
309     void *, pm_bus_power_op_t, void *, void *);
310 static int pm_busop_match_request(dev_info_t *, void *);
311 static int pm_all_to_normal_nexus(dev_info_t *, pm_canblock_t);
312 
313 /*
314  * Dependency Processing is done thru a seperate thread.
315  */
316 kmutex_t	pm_dep_thread_lock;
317 kcondvar_t	pm_dep_thread_cv;
318 pm_dep_wk_t	*pm_dep_thread_workq = NULL;
319 pm_dep_wk_t	*pm_dep_thread_tail = NULL;
320 
321 /*
322  * Autopm  must be turned on by a PM_START_PM ioctl, so we don't end up
323  * power managing things in single user mode that have been suppressed via
324  * power.conf entries.  Protected by pm_scan_lock.
325  */
326 int		autopm_enabled;
327 
328 /*
329  * This flag is true while processes are stopped for a checkpoint/resume.
330  * Controlling processes of direct pm'd devices are not available to
331  * participate in power level changes, so we bypass them when this is set.
332  */
333 static int	pm_processes_stopped;
334 
335 #ifdef	DEBUG
336 
337 /*
338  * see common/sys/epm.h for PMD_* values
339  */
340 uint_t		pm_debug = 0;
341 
342 /*
343  * If pm_divertdebug is set, then no prom_printf calls will be made by
344  * PMD(), which will prevent debug output from bringing up the console
345  * frame buffer.  Clearing this variable before setting pm_debug will result
346  * in PMD output going to the console.
347  *
348  * pm_divertdebug is incremented in pm_set_power() if dip == cfb_dip to avoid
349  * deadlocks and decremented at the end of pm_set_power()
350  */
351 uint_t		pm_divertdebug = 1;
352 kmutex_t	pm_debug_lock;		/* protects pm_divertdebug */
353 
354 void prdeps(char *);
355 #endif
356 
357 /* Globals */
358 
359 /*
360  * List of recorded thresholds and dependencies
361  */
362 pm_thresh_rec_t *pm_thresh_head;
363 krwlock_t pm_thresh_rwlock;
364 
365 pm_pdr_t *pm_dep_head;
366 static int pm_unresolved_deps = 0;
367 static int pm_prop_deps = 0;
368 
369 /*
370  * List of devices that exported no-involuntary-power-cycles property
371  */
372 pm_noinvol_t *pm_noinvol_head;
373 
374 /*
375  * Locks used in noinvol processing
376  */
377 krwlock_t pm_noinvol_rwlock;
378 kmutex_t pm_remdrv_lock;
379 
380 int pm_default_idle_threshold = PM_DEFAULT_SYS_IDLENESS;
381 int pm_system_idle_threshold;
382 /*
383  * By default nexus has 0 threshold, and depends on its children to keep it up
384  */
385 int pm_default_nexus_threshold = 0;
386 
387 /*
388  * Data structures shared with common/io/pm.c
389  */
390 kmutex_t	pm_clone_lock;
391 kcondvar_t	pm_clones_cv[PM_MAX_CLONE];
392 uint_t		pm_poll_cnt[PM_MAX_CLONE];	/* count of events for poll */
393 unsigned char	pm_interest[PM_MAX_CLONE];
394 struct pollhead	pm_pollhead;
395 
396 extern int	hz;
397 extern char	*platform_module_list[];
398 
399 /*
400  * Wrappers for use in ddi_walk_devs
401  */
402 
403 static int		pm_set_dev_thr_walk(dev_info_t *, void *);
404 static int		pm_restore_direct_lvl_walk(dev_info_t *, void *);
405 static int		pm_save_direct_lvl_walk(dev_info_t *, void *);
406 static int		pm_discard_dep_walk(dev_info_t *, void *);
407 #ifdef DEBUG
408 static int		pm_desc_pwrchk_walk(dev_info_t *, void *);
409 #endif
410 
411 /*
412  * Routines for managing noinvol devices
413  */
414 int			pm_noinvol_update(int, int, int, char *, dev_info_t *);
415 void			pm_noinvol_update_node(dev_info_t *,
416 			    pm_bp_noinvol_t *req);
417 
418 kmutex_t pm_rsvp_lock;
419 kmutex_t pm_compcnt_lock;
420 krwlock_t pm_pscc_direct_rwlock;
421 krwlock_t pm_pscc_interest_rwlock;
422 
423 #define	PSC_INTEREST	0	/* belongs to interest psc list */
424 #define	PSC_DIRECT	1	/* belongs to direct psc list */
425 
426 pscc_t *pm_pscc_interest;
427 pscc_t *pm_pscc_direct;
428 
429 #define	PM_MAJOR(dip) ddi_name_to_major(ddi_binding_name(dip))
430 #define	PM_IS_NEXUS(dip) NEXUS_DRV(devopsp[PM_MAJOR(dip)])
431 #define	POWERING_ON(old, new) ((old) == 0 && (new) != 0)
432 #define	POWERING_OFF(old, new) ((old) != 0 && (new) == 0)
433 #define	PPM(dip) ((dev_info_t *)DEVI(dip)->devi_pm_ppm)
434 
435 #define	PM_INCR_NOTLOWEST(dip) {					\
436 	mutex_enter(&pm_compcnt_lock);					\
437 	if (!PM_IS_NEXUS(dip) ||					\
438 	    (DEVI(dip)->devi_pm_flags & (PMC_DEV_THRESH|PMC_COMP_THRESH))) {\
439 		if (pm_comps_notlowest == 0)				\
440 			pm_ppm_notify_all_lowest(dip, PM_NOT_ALL_LOWEST);\
441 		pm_comps_notlowest++;					\
442 		PMD(PMD_LEVEL, ("%s: %s@%s(%s#%d) incr notlowest->%d\n",\
443 		    pmf, PM_DEVICE(dip), pm_comps_notlowest))		\
444 	}								\
445 	mutex_exit(&pm_compcnt_lock);					\
446 }
447 #define	PM_DECR_NOTLOWEST(dip) {					\
448 	mutex_enter(&pm_compcnt_lock);					\
449 	if (!PM_IS_NEXUS(dip) ||					\
450 	    (DEVI(dip)->devi_pm_flags & (PMC_DEV_THRESH|PMC_COMP_THRESH))) {\
451 		ASSERT(pm_comps_notlowest);				\
452 		pm_comps_notlowest--;					\
453 		PMD(PMD_LEVEL, ("%s: %s@%s(%s#%d) decr notlowest to "	\
454 			    "%d\n", pmf, PM_DEVICE(dip), pm_comps_notlowest))\
455 		if (pm_comps_notlowest == 0)				\
456 			pm_ppm_notify_all_lowest(dip, PM_ALL_LOWEST);	\
457 	}								\
458 	mutex_exit(&pm_compcnt_lock);					\
459 }
460 
461 /*
462  * console frame-buffer power-management is not enabled when
463  * debugging services are present.  to override, set pm_cfb_override
464  * to non-zero.
465  */
466 uint_t pm_cfb_comps_off = 0;	/* PM_LEVEL_UNKNOWN is considered on */
467 kmutex_t pm_cfb_lock;
468 int pm_cfb_enabled = 1;		/* non-zero allows pm of console frame buffer */
469 #ifdef DEBUG
470 int pm_cfb_override = 1;	/* non-zero allows pm of cfb with debuggers */
471 #else
472 int pm_cfb_override = 0;	/* non-zero allows pm of cfb with debuggers */
473 #endif
474 
475 static dev_info_t *cfb_dip = 0;
476 static dev_info_t *cfb_dip_detaching = 0;
477 uint_t cfb_inuse = 0;
478 static ddi_softintr_t pm_soft_id;
479 static clock_t pm_soft_pending;
480 int	pm_scans_disabled = 0;
481 
482 /*
483  * A structure to record the fact that one thread has borrowed a lock held
484  * by another thread.  The context requires that the lender block on the
485  * completion of the borrower.
486  */
487 typedef struct lock_loan {
488 	struct lock_loan	*pmlk_next;
489 	kthread_t		*pmlk_borrower;
490 	kthread_t		*pmlk_lender;
491 	dev_info_t		*pmlk_dip;
492 } lock_loan_t;
493 static lock_loan_t lock_loan_head;	/* list head is a dummy element */
494 
495 #ifdef	DEBUG
496 #define	PMD_FUNC(func, name)	char *(func) = (name);
497 #else
498 #define	PMD_FUNC(func, name)
499 #endif
500 
501 
502 /*
503  * Must be called before first device (including pseudo) attach
504  */
505 void
506 pm_init_locks(void)
507 {
508 	mutex_init(&pm_scan_lock, NULL, MUTEX_DRIVER, NULL);
509 	mutex_init(&pm_rsvp_lock, NULL, MUTEX_DRIVER, NULL);
510 	mutex_init(&pm_compcnt_lock, NULL, MUTEX_DRIVER, NULL);
511 	mutex_init(&pm_dep_thread_lock, NULL, MUTEX_DRIVER, NULL);
512 	mutex_init(&pm_remdrv_lock, NULL, MUTEX_DRIVER, NULL);
513 	mutex_init(&pm_loan_lock, NULL, MUTEX_DRIVER, NULL);
514 	rw_init(&pm_thresh_rwlock, NULL, RW_DEFAULT, NULL);
515 	rw_init(&pm_noinvol_rwlock, NULL, RW_DEFAULT, NULL);
516 	cv_init(&pm_dep_thread_cv, NULL, CV_DEFAULT, NULL);
517 }
518 
519 static boolean_t
520 pm_cpr_callb(void *arg, int code)
521 {
522 	_NOTE(ARGUNUSED(arg))
523 	static int auto_save;
524 	static int pm_reset_timestamps(dev_info_t *, void *);
525 
526 	switch (code) {
527 	case CB_CODE_CPR_CHKPT:
528 		/*
529 		 * Cancel scan or wait for scan in progress to finish
530 		 * Other threads may be trying to restart the scan, so we
531 		 * have to keep at it unil it sticks
532 		 */
533 		mutex_enter(&pm_scan_lock);
534 		ASSERT(!pm_scans_disabled);
535 		pm_scans_disabled = 1;
536 		auto_save = autopm_enabled;
537 		autopm_enabled = 0;
538 		mutex_exit(&pm_scan_lock);
539 		ddi_walk_devs(ddi_root_node(), pm_scan_stop_walk, NULL);
540 		break;
541 
542 	case CB_CODE_CPR_RESUME:
543 		ASSERT(!autopm_enabled);
544 		ASSERT(pm_scans_disabled);
545 		pm_scans_disabled = 0;
546 		/*
547 		 * Call pm_reset_timestamps to reset timestamps of each
548 		 * device to the time when the system is resumed so that their
549 		 * idleness can be re-calculated. That's to avoid devices from
550 		 * being powered down right after resume if the system was in
551 		 * suspended mode long enough.
552 		 */
553 		ddi_walk_devs(ddi_root_node(), pm_reset_timestamps, NULL);
554 
555 		autopm_enabled = auto_save;
556 		/*
557 		 * If there is any auto-pm device, get the scanning
558 		 * going. Otherwise don't bother.
559 		 */
560 		ddi_walk_devs(ddi_root_node(), pm_rescan_walk, NULL);
561 		break;
562 	}
563 	return (B_TRUE);
564 }
565 
566 /*
567  * This callback routine is called when there is a system panic.  This function
568  * exists for prototype matching.
569  */
570 static boolean_t
571 pm_panic_callb(void *arg, int code)
572 {
573 	_NOTE(ARGUNUSED(arg, code))
574 	void pm_cfb_check_and_powerup(void);
575 	PMD(PMD_CFB, ("pm_panic_callb\n"))
576 	pm_cfb_check_and_powerup();
577 	return (B_TRUE);
578 }
579 
580 static boolean_t
581 pm_halt_callb(void *arg, int code)
582 {
583 	_NOTE(ARGUNUSED(arg, code))
584 	return (B_TRUE);	/* XXX for now */
585 }
586 
587 /*
588  * This needs to be called after the root and platform drivers are loaded
589  * and be single-threaded with respect to driver attach/detach
590  */
591 void
592 pm_init(void)
593 {
594 	PMD_FUNC(pmf, "pm_init")
595 	char **mod;
596 	extern pri_t minclsyspri;
597 	static void pm_dep_thread(void);
598 
599 	pm_comps_notlowest = 0;
600 	pm_system_idle_threshold = pm_default_idle_threshold;
601 
602 	pm_cpr_cb_id = callb_add(pm_cpr_callb, (void *)NULL,
603 	    CB_CL_CPR_PM, "pm_cpr");
604 	pm_panic_cb_id = callb_add(pm_panic_callb, (void *)NULL,
605 		    CB_CL_PANIC, "pm_panic");
606 	pm_halt_cb_id = callb_add(pm_halt_callb, (void *)NULL,
607 		    CB_CL_HALT, "pm_halt");
608 
609 	/*
610 	 * Create a thread to do dependency processing.
611 	 */
612 	(void) thread_create(NULL, 0, (void (*)())pm_dep_thread, NULL, 0, &p0,
613 	    TS_RUN, minclsyspri);
614 
615 	/*
616 	 * loadrootmodules already loaded these ppm drivers, now get them
617 	 * attached so they can claim the root drivers as they attach
618 	 */
619 	for (mod = platform_module_list; *mod; mod++) {
620 		if (i_ddi_attach_hw_nodes(*mod) != DDI_SUCCESS) {
621 			cmn_err(CE_WARN, "!cannot load platform pm driver %s\n",
622 			    *mod);
623 		} else {
624 			PMD(PMD_DHR, ("%s: %s (%s)\n", pmf, *mod,
625 			    ddi_major_to_name(ddi_name_to_major(*mod))))
626 		}
627 	}
628 }
629 
630 /*
631  * pm_scan_init - create pm scan data structure.  Called (if autopm enabled)
632  * when device becomes power managed or after a failed detach and when autopm
633  * is started via PM_START_PM ioctl, and after a CPR resume to get all the
634  * devices scanning again.
635  */
636 void
637 pm_scan_init(dev_info_t *dip)
638 {
639 	PMD_FUNC(pmf, "scan_init")
640 	pm_scan_t	*scanp;
641 
642 	ASSERT(!PM_ISBC(dip));
643 
644 	PM_LOCK_DIP(dip);
645 	scanp = PM_GET_PM_SCAN(dip);
646 	if (!scanp) {
647 		PMD(PMD_SCAN, ("%s: %s@%s(%s#%d): create scan data\n",
648 		    pmf, PM_DEVICE(dip)))
649 		scanp =  kmem_zalloc(sizeof (pm_scan_t), KM_SLEEP);
650 		DEVI(dip)->devi_pm_scan = scanp;
651 	} else if (scanp->ps_scan_flags & PM_SCAN_STOP) {
652 		PMD(PMD_SCAN, ("%s: %s@%s(%s#%d): "
653 		    "clear PM_SCAN_STOP flag\n", pmf, PM_DEVICE(dip)))
654 		scanp->ps_scan_flags &= ~PM_SCAN_STOP;
655 	}
656 	PM_UNLOCK_DIP(dip);
657 }
658 
659 /*
660  * pm_scan_fini - remove pm scan data structure when stopping pm on the device
661  */
662 void
663 pm_scan_fini(dev_info_t *dip)
664 {
665 	PMD_FUNC(pmf, "scan_fini")
666 	pm_scan_t	*scanp;
667 
668 	PMD(PMD_SCAN, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
669 	ASSERT(!PM_ISBC(dip));
670 	PM_LOCK_DIP(dip);
671 	scanp = PM_GET_PM_SCAN(dip);
672 	if (!scanp) {
673 		PM_UNLOCK_DIP(dip);
674 		return;
675 	}
676 
677 	ASSERT(!scanp->ps_scan_id && !(scanp->ps_scan_flags &
678 	    (PM_SCANNING | PM_SCAN_DISPATCHED | PM_SCAN_AGAIN)));
679 
680 	kmem_free(scanp, sizeof (pm_scan_t));
681 	DEVI(dip)->devi_pm_scan = NULL;
682 	PM_UNLOCK_DIP(dip);
683 }
684 
685 /*
686  * Given a pointer to a component struct, return the current power level
687  * (struct contains index unless it is a continuous level).
688  * Located here in hopes of getting both this and dev_is_needed into the
689  * cache together
690  */
691 static int
692 cur_power(pm_component_t *cp)
693 {
694 	if (cp->pmc_cur_pwr == PM_LEVEL_UNKNOWN)
695 		return (cp->pmc_cur_pwr);
696 
697 	return (cp->pmc_comp.pmc_lvals[cp->pmc_cur_pwr]);
698 }
699 
700 static char *
701 pm_decode_direction(int direction)
702 {
703 	switch (direction) {
704 	case PM_LEVEL_UPONLY:
705 		return ("up");
706 
707 	case PM_LEVEL_EXACT:
708 		return ("exact");
709 
710 	case PM_LEVEL_DOWNONLY:
711 		return ("down");
712 
713 	default:
714 		return ("INVALID DIRECTION");
715 	}
716 }
717 
718 char *
719 pm_decode_op(pm_bus_power_op_t op)
720 {
721 	switch (op) {
722 	case BUS_POWER_CHILD_PWRCHG:
723 		return ("CHILD_PWRCHG");
724 	case BUS_POWER_NEXUS_PWRUP:
725 		return ("NEXUS_PWRUP");
726 	case BUS_POWER_PRE_NOTIFICATION:
727 		return ("PRE_NOTIFICATION");
728 	case BUS_POWER_POST_NOTIFICATION:
729 		return ("POST_NOTIFICATION");
730 	case BUS_POWER_HAS_CHANGED:
731 		return ("HAS_CHANGED");
732 	case BUS_POWER_NOINVOL:
733 		return ("NOINVOL");
734 	default:
735 		return ("UNKNOWN OP");
736 	}
737 }
738 
739 /*
740  * Returns true if level is a possible (valid) power level for component
741  */
742 int
743 e_pm_valid_power(dev_info_t *dip, int cmpt, int level)
744 {
745 	PMD_FUNC(pmf, "e_pm_valid_power")
746 	pm_component_t *cp = PM_CP(dip, cmpt);
747 	int i;
748 	int *ip = cp->pmc_comp.pmc_lvals;
749 	int limit = cp->pmc_comp.pmc_numlevels;
750 
751 	if (level < 0)
752 		return (0);
753 	for (i = 0; i < limit; i++) {
754 		if (level == *ip++)
755 			return (1);
756 	}
757 #ifdef DEBUG
758 	if (pm_debug & PMD_FAIL) {
759 		ip = cp->pmc_comp.pmc_lvals;
760 
761 		for (i = 0; i < limit; i++)
762 			PMD(PMD_FAIL, ("%s: index=%d, level=%d\n",
763 			    pmf, i, *ip++))
764 	}
765 #endif
766 	return (0);
767 }
768 
769 /*
770  * Returns true if device is pm'd (after calling pm_start if need be)
771  */
772 int
773 e_pm_valid_info(dev_info_t *dip, pm_info_t **infop)
774 {
775 	pm_info_t *info;
776 	static int pm_start(dev_info_t *dip);
777 
778 	/*
779 	 * Check if the device is power managed if not.
780 	 * To make the common case (device is power managed already)
781 	 * fast, we check without the lock.  If device is not already
782 	 * power managed, then we take the lock and the long route through
783 	 * go get it managed.  Devices never go unmanaged until they
784 	 * detach.
785 	 */
786 	info = PM_GET_PM_INFO(dip);
787 	if (!info) {
788 		if (!DEVI_IS_ATTACHING(dip)) {
789 			return (0);
790 		}
791 		if (pm_start(dip) != DDI_SUCCESS) {
792 			return (0);
793 		}
794 		info = PM_GET_PM_INFO(dip);
795 	}
796 	ASSERT(info);
797 	if (infop != NULL)
798 		*infop = info;
799 	return (1);
800 }
801 
802 int
803 e_pm_valid_comp(dev_info_t *dip, int cmpt, pm_component_t **cpp)
804 {
805 	if (cmpt >= 0 && cmpt < PM_NUMCMPTS(dip)) {
806 		if (cpp != NULL)
807 			*cpp = PM_CP(dip, cmpt);
808 		return (1);
809 	} else {
810 		return (0);
811 	}
812 }
813 
814 /*
815  * Internal guts of ddi_dev_is_needed and pm_raise/lower_power
816  */
817 static int
818 dev_is_needed(dev_info_t *dip, int cmpt, int level, int direction)
819 {
820 	PMD_FUNC(pmf, "din")
821 	pm_component_t *cp;
822 	char *pathbuf;
823 	int result;
824 
825 	ASSERT(direction == PM_LEVEL_UPONLY || direction == PM_LEVEL_DOWNONLY);
826 	if (!e_pm_valid_info(dip, NULL) || !e_pm_valid_comp(dip, cmpt, &cp) ||
827 	    !e_pm_valid_power(dip, cmpt, level))
828 		return (DDI_FAILURE);
829 
830 	PMD(PMD_DIN, ("%s: %s@%s(%s#%d) cmpt=%d, dir=%s, new=%d, cur=%d\n",
831 	    pmf, PM_DEVICE(dip), cmpt, pm_decode_direction(direction),
832 	    level, cur_power(cp)))
833 
834 	if (pm_set_power(dip, cmpt, level,  direction,
835 	    PM_CANBLOCK_BLOCK, 0, &result) != DDI_SUCCESS) {
836 		if (direction == PM_LEVEL_UPONLY) {
837 			pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
838 			(void) ddi_pathname(dip, pathbuf);
839 			cmn_err(CE_WARN, "Device %s failed to power up.",
840 			    pathbuf);
841 			kmem_free(pathbuf, MAXPATHLEN);
842 		}
843 		PMD(PMD_DIN | PMD_FAIL, ("%s: %s@%s(%s#%d) [%d] %s->%d failed, "
844 		    "errno %d\n", pmf, PM_DEVICE(dip), cmpt,
845 		    pm_decode_direction(direction), level, result))
846 		return (DDI_FAILURE);
847 	}
848 
849 	PMD(PMD_RESCAN | PMD_DIN, ("%s: pm_rescan %s@%s(%s#%d)\n", pmf,
850 	    PM_DEVICE(dip)))
851 	pm_rescan(dip);
852 	return (DDI_SUCCESS);
853 }
854 
855 /*
856  * We can get multiple pm_rescan() threads, if one of them discovers
857  * that no scan is running at the moment, it kicks it into action.
858  * Otherwise, it tells the current scanning thread to scan again when
859  * it is done by asserting the PM_SCAN_AGAIN flag. The PM_SCANNING and
860  * PM_SCAN_AGAIN flags are used to regulate scan, to make sure only one
861  * thread at a time runs the pm_scan_dev() code.
862  */
863 void
864 pm_rescan(void *arg)
865 {
866 	PMD_FUNC(pmf, "rescan")
867 	dev_info_t	*dip = (dev_info_t *)arg;
868 	pm_info_t	*info;
869 	pm_scan_t	*scanp;
870 	timeout_id_t	scanid;
871 
872 	PMD(PMD_SCAN, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
873 	PM_LOCK_DIP(dip);
874 	info = PM_GET_PM_INFO(dip);
875 	scanp = PM_GET_PM_SCAN(dip);
876 	if (pm_scans_disabled || !autopm_enabled || !info || !scanp ||
877 	    (scanp->ps_scan_flags & PM_SCAN_STOP)) {
878 		PM_UNLOCK_DIP(dip);
879 		return;
880 	}
881 	if (scanp->ps_scan_flags & PM_SCANNING) {
882 		scanp->ps_scan_flags |= PM_SCAN_AGAIN;
883 		PM_UNLOCK_DIP(dip);
884 		return;
885 	} else if (scanp->ps_scan_id) {
886 		scanid = scanp->ps_scan_id;
887 		scanp->ps_scan_id = 0;
888 		PMD(PMD_SCAN, ("%s: %s@%s(%s#%d): cancel timeout scanid %lx\n",
889 		    pmf, PM_DEVICE(dip), (ulong_t)scanid))
890 		PM_UNLOCK_DIP(dip);
891 		(void) untimeout(scanid);
892 		PM_LOCK_DIP(dip);
893 	}
894 
895 	/*
896 	 * Dispatching pm_scan during attach time is risky due to the fact that
897 	 * attach might soon fail and dip dissolved, and panic may happen while
898 	 * attempting to stop scan. So schedule a pm_rescan instead.
899 	 * (Note that if either of the first two terms are true, taskq_dispatch
900 	 * will not be invoked).
901 	 *
902 	 * Multiple pm_scan dispatching is unecessary and costly to keep track
903 	 * of. The PM_SCAN_DISPATCHED flag is used between pm_rescan and pm_scan
904 	 * to regulate the dispatching.
905 	 *
906 	 * Scan is stopped before the device is detached (in pm_detaching())
907 	 * but it may get re-started during the post_detach processing if the
908 	 * driver fails to detach.
909 	 */
910 	if (DEVI_IS_ATTACHING(dip) ||
911 	    (scanp->ps_scan_flags & PM_SCAN_DISPATCHED) ||
912 	    !taskq_dispatch(system_taskq, pm_scan, (void *)dip, TQ_NOSLEEP)) {
913 		PMD(PMD_SCAN, ("%s: %s@%s(%s#%d): attaching, pm_scan already "
914 		    "dispatched or dispatching failed\n", pmf, PM_DEVICE(dip)))
915 		if (scanp->ps_scan_id) {
916 			scanid = scanp->ps_scan_id;
917 			scanp->ps_scan_id = 0;
918 			PM_UNLOCK_DIP(dip);
919 			(void) untimeout(scanid);
920 			PM_LOCK_DIP(dip);
921 			if (scanp->ps_scan_id) {
922 				PMD(PMD_SCAN, ("%s: %s@%s(%s#%d): a competing "
923 				    "thread scheduled pm_rescan, scanid %lx\n",
924 				    pmf, PM_DEVICE(dip),
925 				    (ulong_t)scanp->ps_scan_id))
926 				PM_UNLOCK_DIP(dip);
927 				return;
928 			}
929 		}
930 		scanp->ps_scan_id = timeout(pm_rescan, (void *)dip,
931 		    (scanp->ps_idle_down ? pm_id_ticks :
932 		    (pm_min_scan * hz)));
933 		PMD(PMD_SCAN, ("%s: %s@%s(%s#%d): scheduled next pm_rescan, "
934 		    "scanid %lx\n", pmf, PM_DEVICE(dip),
935 		    (ulong_t)scanp->ps_scan_id))
936 	} else {
937 		PMD(PMD_SCAN, ("%s: dispatched pm_scan for %s@%s(%s#%d)\n",
938 		    pmf, PM_DEVICE(dip)))
939 		scanp->ps_scan_flags |= PM_SCAN_DISPATCHED;
940 	}
941 	PM_UNLOCK_DIP(dip);
942 }
943 
944 void
945 pm_scan(void *arg)
946 {
947 	PMD_FUNC(pmf, "scan")
948 	dev_info_t	*dip = (dev_info_t *)arg;
949 	pm_scan_t	*scanp;
950 	time_t		nextscan;
951 
952 	PMD(PMD_SCAN, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
953 
954 	PM_LOCK_DIP(dip);
955 	scanp = PM_GET_PM_SCAN(dip);
956 	ASSERT(scanp && PM_GET_PM_INFO(dip));
957 
958 	if (pm_scans_disabled || !autopm_enabled ||
959 	    (scanp->ps_scan_flags & PM_SCAN_STOP)) {
960 		scanp->ps_scan_flags &= ~(PM_SCAN_AGAIN | PM_SCAN_DISPATCHED);
961 		PM_UNLOCK_DIP(dip);
962 		return;
963 	}
964 
965 	if (scanp->ps_idle_down) {
966 		/*
967 		 * make sure we remember idledown was in affect until
968 		 * we've completed the scan
969 		 */
970 		PMID_SET_SCANS(scanp->ps_idle_down)
971 		PMD(PMD_IDLEDOWN, ("%s: %s@%s(%s#%d): idledown starts "
972 		    "(pmid %x)\n", pmf, PM_DEVICE(dip), scanp->ps_idle_down))
973 	}
974 
975 	/* possible having two threads running pm_scan() */
976 	if (scanp->ps_scan_flags & PM_SCANNING) {
977 		scanp->ps_scan_flags |= PM_SCAN_AGAIN;
978 		PMD(PMD_SCAN, ("%s: scanning, will scan %s@%s(%s#%d) again\n",
979 		    pmf, PM_DEVICE(dip)))
980 		scanp->ps_scan_flags &= ~PM_SCAN_DISPATCHED;
981 		PM_UNLOCK_DIP(dip);
982 		return;
983 	}
984 
985 	scanp->ps_scan_flags |= PM_SCANNING;
986 	scanp->ps_scan_flags &= ~PM_SCAN_DISPATCHED;
987 	do {
988 		scanp->ps_scan_flags &= ~PM_SCAN_AGAIN;
989 		PM_UNLOCK_DIP(dip);
990 		nextscan = pm_scan_dev(dip);
991 		PM_LOCK_DIP(dip);
992 	} while (scanp->ps_scan_flags & PM_SCAN_AGAIN);
993 
994 	ASSERT(scanp->ps_scan_flags & PM_SCANNING);
995 	scanp->ps_scan_flags &= ~PM_SCANNING;
996 
997 	if (scanp->ps_idle_down) {
998 		scanp->ps_idle_down &= ~PMID_SCANS;
999 		PMD(PMD_IDLEDOWN, ("%s: %s@%s(%s#%d): idledown ends "
1000 		    "(pmid %x)\n", pmf, PM_DEVICE(dip), scanp->ps_idle_down))
1001 	}
1002 
1003 	/* schedule for next idle check */
1004 	if (nextscan != LONG_MAX) {
1005 		if (nextscan > (LONG_MAX / hz))
1006 			nextscan = (LONG_MAX - 1) / hz;
1007 		if (scanp->ps_scan_id) {
1008 			PMD(PMD_SCAN, ("%s: %s@%s(%s#%d): while scanning "
1009 			    "another rescan scheduled scanid(%lx)\n", pmf,
1010 			    PM_DEVICE(dip), (ulong_t)scanp->ps_scan_id))
1011 			PM_UNLOCK_DIP(dip);
1012 			return;
1013 		} else if (!(scanp->ps_scan_flags & PM_SCAN_STOP)) {
1014 			scanp->ps_scan_id = timeout(pm_rescan, (void *)dip,
1015 			    (clock_t)(nextscan * hz));
1016 			PMD(PMD_SCAN, ("%s: nextscan for %s@%s(%s#%d) in "
1017 			    "%lx sec, scanid(%lx) \n", pmf, PM_DEVICE(dip),
1018 			    (ulong_t)nextscan, (ulong_t)scanp->ps_scan_id))
1019 		}
1020 	}
1021 	PM_UNLOCK_DIP(dip);
1022 }
1023 
1024 void
1025 pm_get_timestamps(dev_info_t *dip, time_t *valuep)
1026 {
1027 	int components = PM_NUMCMPTS(dip);
1028 	int i;
1029 
1030 	ASSERT(components > 0);
1031 	PM_LOCK_BUSY(dip);	/* so we get a consistent view */
1032 	for (i = 0; i < components; i++) {
1033 		valuep[i] = PM_CP(dip, i)->pmc_timestamp;
1034 	}
1035 	PM_UNLOCK_BUSY(dip);
1036 }
1037 
1038 /*
1039  * Returns true if device needs to be kept up because it exported the
1040  * "no-involuntary-power-cycles" property or we're pretending it did (console
1041  * fb case) or it is an ancestor of such a device and has used up the "one
1042  * free cycle" allowed when all such leaf nodes have voluntarily powered down
1043  * upon detach
1044  */
1045 int
1046 pm_noinvol(dev_info_t *dip)
1047 {
1048 	PMD_FUNC(pmf, "noinvol")
1049 
1050 	/*
1051 	 * This doesn't change over the life of a driver, so no locking needed
1052 	 */
1053 	if (PM_IS_CFB(dip)) {
1054 		PMD(PMD_NOINVOL | PMD_CFB, ("%s: inhibits CFB %s@%s(%s#%d)\n",
1055 		    pmf, PM_DEVICE(dip)))
1056 		return (1);
1057 	}
1058 	/*
1059 	 * Not an issue if no such kids
1060 	 */
1061 	if (DEVI(dip)->devi_pm_noinvolpm == 0) {
1062 #ifdef DEBUG
1063 		if (DEVI(dip)->devi_pm_volpmd != 0) {
1064 			dev_info_t *pdip = dip;
1065 			do {
1066 				PMD(PMD_NOINVOL, ("%s: %s@%s(%s#%d) noinvol %d "
1067 				    "volpmd %d\n", pmf, PM_DEVICE(pdip),
1068 				    DEVI(pdip)->devi_pm_noinvolpm,
1069 				    DEVI(pdip)->devi_pm_volpmd))
1070 				pdip = ddi_get_parent(pdip);
1071 			} while (pdip);
1072 		}
1073 #endif
1074 		ASSERT(DEVI(dip)->devi_pm_volpmd == 0);
1075 		return (0);
1076 	}
1077 
1078 	/*
1079 	 * Since we now maintain the counts correct at every node, we no longer
1080 	 * need to look up the tree.  An ancestor cannot use up the free cycle
1081 	 * without the children getting their counts adjusted.
1082 	 */
1083 
1084 #ifdef	DEBUG
1085 	if (DEVI(dip)->devi_pm_noinvolpm != DEVI(dip)->devi_pm_volpmd)
1086 		PMD(PMD_NOINVOL, ("%s: (%d != %d) inhibits %s@%s(%s#%d)\n", pmf,
1087 		    DEVI(dip)->devi_pm_noinvolpm, DEVI(dip)->devi_pm_volpmd,
1088 		    PM_DEVICE(dip)))
1089 #endif
1090 	return (DEVI(dip)->devi_pm_noinvolpm != DEVI(dip)->devi_pm_volpmd);
1091 }
1092 
1093 /*
1094  * This function performs the actual scanning of the device.
1095  * It attempts to power off the indicated device's components if they have
1096  * been idle and other restrictions are met.
1097  * pm_scan_dev calculates and returns when the next scan should happen for
1098  * this device.
1099  */
1100 time_t
1101 pm_scan_dev(dev_info_t *dip)
1102 {
1103 	PMD_FUNC(pmf, "scan_dev")
1104 	pm_scan_t	*scanp;
1105 	time_t		*timestamp, idletime, now, thresh;
1106 	time_t		timeleft = 0;
1107 	int		i, nxtpwr, curpwr, pwrndx, unused;
1108 	size_t		size;
1109 	pm_component_t	 *cp;
1110 	dev_info_t	*pdip = ddi_get_parent(dip);
1111 	int		circ;
1112 	static int	cur_threshold(dev_info_t *, int);
1113 	static int	pm_next_lower_power(pm_component_t *, int);
1114 
1115 	/*
1116 	 * skip attaching device
1117 	 */
1118 	if (DEVI_IS_ATTACHING(dip)) {
1119 		PMD(PMD_SCAN, ("%s: %s@%s(%s#%d) is attaching, timeleft(%lx)\n",
1120 		    pmf, PM_DEVICE(dip), pm_min_scan))
1121 		return (pm_min_scan);
1122 	}
1123 
1124 	PM_LOCK_DIP(dip);
1125 	scanp = PM_GET_PM_SCAN(dip);
1126 	ASSERT(scanp && PM_GET_PM_INFO(dip));
1127 
1128 	PMD(PMD_SCAN, ("%s: [BEGIN %s@%s(%s#%d)]\n", pmf, PM_DEVICE(dip)))
1129 	PMD(PMD_SCAN, ("%s: %s@%s(%s#%d): kuc is %d\n", pmf, PM_DEVICE(dip),
1130 	    PM_KUC(dip)))
1131 
1132 	/* no scan under the following conditions */
1133 	if (pm_scans_disabled || !autopm_enabled ||
1134 	    (scanp->ps_scan_flags & PM_SCAN_STOP) ||
1135 	    (PM_KUC(dip) != 0) ||
1136 	    PM_ISDIRECT(dip) || pm_noinvol(dip)) {
1137 		PM_UNLOCK_DIP(dip);
1138 		PMD(PMD_SCAN, ("%s: [END, %s@%s(%s#%d)] no scan, "
1139 		    "scan_disabled(%d), apm_enabled(%d), kuc(%d), "
1140 		    "%s directpm, %s pm_noinvol\n", pmf, PM_DEVICE(dip),
1141 		    pm_scans_disabled, autopm_enabled, PM_KUC(dip),
1142 		    PM_ISDIRECT(dip) ? "is" : "is not",
1143 		    pm_noinvol(dip) ? "is" : "is not"))
1144 		return (LONG_MAX);
1145 	}
1146 	PM_UNLOCK_DIP(dip);
1147 
1148 	if (!ndi_devi_tryenter(pdip, &circ)) {
1149 		PMD(PMD_SCAN, ("%s: %s@%s(%s#%d) can't hold pdip",
1150 		    pmf, PM_DEVICE(pdip)))
1151 		return ((time_t)1);
1152 	}
1153 	now = gethrestime_sec();
1154 	size = PM_NUMCMPTS(dip) * sizeof (time_t);
1155 	timestamp = kmem_alloc(size, KM_SLEEP);
1156 	pm_get_timestamps(dip, timestamp);
1157 
1158 	/*
1159 	 * Since we removed support for backwards compatible devices,
1160 	 * (see big comment at top of file)
1161 	 * it is no longer required to deal with component 0 last.
1162 	 */
1163 	for (i = 0; i < PM_NUMCMPTS(dip); i++) {
1164 		/*
1165 		 * If already off (an optimization, perhaps)
1166 		 */
1167 		cp = PM_CP(dip, i);
1168 		pwrndx = cp->pmc_cur_pwr;
1169 		curpwr = (pwrndx == PM_LEVEL_UNKNOWN) ?
1170 		    PM_LEVEL_UNKNOWN :
1171 		    cp->pmc_comp.pmc_lvals[pwrndx];
1172 
1173 		if (pwrndx == 0) {
1174 			PMD(PMD_SCAN, ("%s: %s@%s(%s#%d) comp %d off or "
1175 			    "lowest\n", pmf, PM_DEVICE(dip), i))
1176 			/* skip device if off or at its lowest */
1177 			continue;
1178 		}
1179 
1180 		thresh = cur_threshold(dip, i);		/* comp i threshold */
1181 		if ((timestamp[i] == 0) || (cp->pmc_busycount > 0)) {
1182 			/* were busy or newly became busy by another thread */
1183 			if (timeleft == 0)
1184 				timeleft = max(thresh, pm_min_scan);
1185 			else
1186 				timeleft = min(
1187 				    timeleft, max(thresh, pm_min_scan));
1188 			continue;
1189 		}
1190 
1191 		idletime = now - timestamp[i];		/* idle time */
1192 		PMD(PMD_SCAN, ("%s: %s@%s(%s#%d) comp %d idle time %lx\n",
1193 		    pmf, PM_DEVICE(dip), i, idletime))
1194 		if (idletime >= thresh || PM_IS_PID(dip)) {
1195 			nxtpwr = pm_next_lower_power(cp, pwrndx);
1196 			PMD(PMD_SCAN, ("%s: %s@%s(%s#%d) comp %d, %d->%d\n",
1197 			    pmf, PM_DEVICE(dip), i, curpwr, nxtpwr))
1198 			if (pm_set_power(dip, i, nxtpwr, PM_LEVEL_DOWNONLY,
1199 			    PM_CANBLOCK_FAIL, 1, &unused) != DDI_SUCCESS &&
1200 			    PM_CURPOWER(dip, i) != nxtpwr) {
1201 				PMD(PMD_SCAN, ("%s: %s@%s(%s#%d) comp %d, "
1202 				    "%d->%d Failed\n", pmf, PM_DEVICE(dip),
1203 				    i, curpwr, nxtpwr))
1204 				timeleft = pm_min_scan;
1205 				continue;
1206 			} else {
1207 				PMD(PMD_SCAN, ("%s: %s@%s(%s#%d) comp %d, "
1208 				    "%d->%d, GOOD curpwr %d\n", pmf,
1209 				    PM_DEVICE(dip), i, curpwr, nxtpwr,
1210 				    cur_power(cp)))
1211 
1212 				if (nxtpwr == 0)	/* component went off */
1213 					continue;
1214 
1215 				/*
1216 				 * scan to next lower level
1217 				 */
1218 				if (timeleft == 0)
1219 					timeleft = max(
1220 					    1, cur_threshold(dip, i));
1221 				else
1222 					timeleft = min(timeleft,
1223 					    max(1, cur_threshold(dip, i)));
1224 				PMD(PMD_SCAN, ("%s: %s@%s(%s#%d) comp %d, "
1225 				    "timeleft(%lx)\n", pmf, PM_DEVICE(dip),
1226 				    i, timeleft))
1227 			}
1228 		} else {	/* comp not idle long enough */
1229 			if (timeleft == 0)
1230 				timeleft = thresh - idletime;
1231 			else
1232 				timeleft = min(timeleft, (thresh - idletime));
1233 			PMD(PMD_SCAN, ("%s: %s@%s(%s#%d) comp %d, timeleft="
1234 			    "%lx\n", pmf, PM_DEVICE(dip), i, timeleft))
1235 		}
1236 	}
1237 	ndi_devi_exit(pdip, circ);
1238 	kmem_free(timestamp, size);
1239 	PMD(PMD_SCAN, ("%s: [END %s@%s(%s#%d)] timeleft(%lx)\n", pmf,
1240 	    PM_DEVICE(dip), timeleft))
1241 
1242 	/*
1243 	 * if components are already at lowest level, timeleft is left 0
1244 	 */
1245 	return ((timeleft == 0) ? LONG_MAX : timeleft);
1246 }
1247 
1248 /*
1249  * pm_scan_stop - cancel scheduled pm_rescan,
1250  *                wait for termination of dispatched pm_scan thread
1251  *                     and active pm_scan_dev thread.
1252  */
1253 void
1254 pm_scan_stop(dev_info_t *dip)
1255 {
1256 	PMD_FUNC(pmf, "scan_stop")
1257 	pm_scan_t	*scanp;
1258 	timeout_id_t	scanid;
1259 
1260 	PMD(PMD_SCAN, ("%s: [BEGIN %s@%s(%s#%d)]\n", pmf, PM_DEVICE(dip)))
1261 	PM_LOCK_DIP(dip);
1262 	scanp = PM_GET_PM_SCAN(dip);
1263 	if (!scanp) {
1264 		PMD(PMD_SCAN, ("%s: [END %s@%s(%s#%d)] scan not initialized\n",
1265 		    pmf, PM_DEVICE(dip)))
1266 		PM_UNLOCK_DIP(dip);
1267 		return;
1268 	}
1269 	scanp->ps_scan_flags |= PM_SCAN_STOP;
1270 
1271 	/* cancel scheduled scan taskq */
1272 	while (scanp->ps_scan_id) {
1273 		scanid = scanp->ps_scan_id;
1274 		scanp->ps_scan_id = 0;
1275 		PM_UNLOCK_DIP(dip);
1276 		(void) untimeout(scanid);
1277 		PM_LOCK_DIP(dip);
1278 	}
1279 
1280 	while (scanp->ps_scan_flags & (PM_SCANNING | PM_SCAN_DISPATCHED)) {
1281 		PM_UNLOCK_DIP(dip);
1282 		delay(1);
1283 		PM_LOCK_DIP(dip);
1284 	}
1285 	PM_UNLOCK_DIP(dip);
1286 	PMD(PMD_SCAN, ("%s: [END %s@%s(%s#%d)]\n", pmf, PM_DEVICE(dip)))
1287 }
1288 
1289 int
1290 pm_scan_stop_walk(dev_info_t *dip, void *arg)
1291 {
1292 	_NOTE(ARGUNUSED(arg))
1293 
1294 	if (!PM_GET_PM_SCAN(dip))
1295 		return (DDI_WALK_CONTINUE);
1296 	ASSERT(!PM_ISBC(dip));
1297 	pm_scan_stop(dip);
1298 	return (DDI_WALK_CONTINUE);
1299 }
1300 
1301 /*
1302  * Converts a power level value to its index
1303  */
1304 static int
1305 power_val_to_index(pm_component_t *cp, int val)
1306 {
1307 	int limit, i, *ip;
1308 
1309 	ASSERT(val != PM_LEVEL_UPONLY && val != PM_LEVEL_DOWNONLY &&
1310 	    val != PM_LEVEL_EXACT);
1311 	/*  convert power value into index (i) */
1312 	limit = cp->pmc_comp.pmc_numlevels;
1313 	ip = cp->pmc_comp.pmc_lvals;
1314 	for (i = 0; i < limit; i++)
1315 		if (val == *ip++)
1316 			return (i);
1317 	return (-1);
1318 }
1319 
1320 /*
1321  * Converts a numeric power level to a printable string
1322  */
1323 static char *
1324 power_val_to_string(pm_component_t *cp, int val)
1325 {
1326 	int index;
1327 
1328 	if (val == PM_LEVEL_UPONLY)
1329 		return ("<UPONLY>");
1330 
1331 	if (val == PM_LEVEL_UNKNOWN ||
1332 	    (index = power_val_to_index(cp, val)) == -1)
1333 		return ("<LEVEL_UNKNOWN>");
1334 
1335 	return (cp->pmc_comp.pmc_lnames[index]);
1336 }
1337 
1338 /*
1339  * Return true if this node has been claimed by a ppm.
1340  */
1341 static int
1342 pm_ppm_claimed(dev_info_t *dip)
1343 {
1344 	return (PPM(dip) != NULL);
1345 }
1346 
1347 /*
1348  * A node which was voluntarily power managed has just used up its "free cycle"
1349  * and need is volpmd field cleared, and the same done to all its descendents
1350  */
1351 static void
1352 pm_clear_volpm_dip(dev_info_t *dip)
1353 {
1354 	PMD_FUNC(pmf, "clear_volpm_dip")
1355 
1356 	if (dip == NULL)
1357 		return;
1358 	PMD(PMD_NOINVOL, ("%s: clear volpm from %s@%s(%s#%d)\n", pmf,
1359 	    PM_DEVICE(dip)))
1360 	DEVI(dip)->devi_pm_volpmd = 0;
1361 	for (dip = ddi_get_child(dip); dip; dip = ddi_get_next_sibling(dip)) {
1362 		pm_clear_volpm_dip(dip);
1363 	}
1364 }
1365 
1366 /*
1367  * A node which was voluntarily power managed has used up the "free cycles"
1368  * for the subtree that it is the root of.  Scan through the list of detached
1369  * nodes and adjust the counts of any that are descendents of the node.
1370  */
1371 static void
1372 pm_clear_volpm_list(dev_info_t *dip)
1373 {
1374 	PMD_FUNC(pmf, "clear_volpm_list")
1375 	char	*pathbuf;
1376 	size_t	len;
1377 	pm_noinvol_t *ip;
1378 
1379 	pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1380 	(void) ddi_pathname(dip, pathbuf);
1381 	len = strlen(pathbuf);
1382 	PMD(PMD_NOINVOL, ("%s: clear volpm list %s\n", pmf, pathbuf))
1383 	rw_enter(&pm_noinvol_rwlock, RW_WRITER);
1384 	for (ip = pm_noinvol_head; ip; ip = ip->ni_next) {
1385 		PMD(PMD_NOINVOL, ("%s: clear volpm: ni_path %s\n", pmf,
1386 		    ip->ni_path))
1387 		if (strncmp(pathbuf, ip->ni_path, len) == 0 &&
1388 		    ip->ni_path[len] == '/') {
1389 			PMD(PMD_NOINVOL, ("%s: clear volpm: %s\n", pmf,
1390 			    ip->ni_path))
1391 			ip->ni_volpmd = 0;
1392 			ip->ni_wasvolpmd = 0;
1393 		}
1394 	}
1395 	kmem_free(pathbuf, MAXPATHLEN);
1396 	rw_exit(&pm_noinvol_rwlock);
1397 }
1398 
1399 /*
1400  * Powers a device, suspending or resuming the driver if it is a backward
1401  * compatible device, calling into ppm to change power level.
1402  * Called with the component's power lock held.
1403  */
1404 static int
1405 power_dev(dev_info_t *dip, int comp, int level, int old_level,
1406     pm_canblock_t canblock, pm_ppm_devlist_t **devlist)
1407 {
1408 	PMD_FUNC(pmf, "power_dev")
1409 	power_req_t power_req;
1410 	int		power_op_ret;	/* DDI_SUCCESS or DDI_FAILURE */
1411 	int		resume_needed = 0;
1412 	int		suspended = 0;
1413 	int		result;
1414 	struct pm_component *cp = PM_CP(dip, comp);
1415 	int		bc = PM_ISBC(dip);
1416 	int pm_all_components_off(dev_info_t *);
1417 	int		clearvolpmd = 0;
1418 	char		pathbuf[MAXNAMELEN];
1419 #ifdef DEBUG
1420 	char *ppmname, *ppmaddr;
1421 #endif
1422 	/*
1423 	 * If this is comp 0 of a backwards compat device and we are
1424 	 * going to take the power away, we need to detach it with
1425 	 * DDI_PM_SUSPEND command.
1426 	 */
1427 	if (bc && comp == 0 && POWERING_OFF(old_level, level)) {
1428 		if (devi_detach(dip, DDI_PM_SUSPEND) != DDI_SUCCESS) {
1429 			/* We could not suspend before turning cmpt zero off */
1430 			PMD(PMD_ERROR, ("%s: could not suspend %s@%s(%s#%d)\n",
1431 			    pmf, PM_DEVICE(dip)))
1432 			return (DDI_FAILURE);
1433 		} else {
1434 			DEVI(dip)->devi_pm_flags |= PMC_SUSPENDED;
1435 			suspended++;
1436 		}
1437 	}
1438 	power_req.request_type = PMR_PPM_SET_POWER;
1439 	power_req.req.ppm_set_power_req.who = dip;
1440 	power_req.req.ppm_set_power_req.cmpt = comp;
1441 	power_req.req.ppm_set_power_req.old_level = old_level;
1442 	power_req.req.ppm_set_power_req.new_level = level;
1443 	power_req.req.ppm_set_power_req.canblock = canblock;
1444 	power_req.req.ppm_set_power_req.cookie = NULL;
1445 #ifdef DEBUG
1446 	if (pm_ppm_claimed(dip)) {
1447 		ppmname = PM_NAME(PPM(dip));
1448 		ppmaddr = PM_ADDR(PPM(dip));
1449 
1450 	} else {
1451 		ppmname = "noppm";
1452 		ppmaddr = "0";
1453 	}
1454 	PMD(PMD_PPM, ("%s: %s@%s(%s#%d):%s[%d] %s (%d) -> %s (%d) via %s@%s\n",
1455 	    pmf, PM_DEVICE(dip), cp->pmc_comp.pmc_name, comp,
1456 	    power_val_to_string(cp, old_level), old_level,
1457 	    power_val_to_string(cp, level), level, ppmname, ppmaddr))
1458 #endif
1459 	/*
1460 	 * If non-bc noinvolpm device is turning first comp on, or noinvolpm
1461 	 * bc device comp 0 is powering on, then we count it as a power cycle
1462 	 * against its voluntary count.
1463 	 */
1464 	if (DEVI(dip)->devi_pm_volpmd &&
1465 	    (!bc && pm_all_components_off(dip) && level != 0) ||
1466 	    (bc && comp == 0 && POWERING_ON(old_level, level)))
1467 		clearvolpmd = 1;
1468 	if ((power_op_ret = pm_ctlops(PPM(dip), dip, DDI_CTLOPS_POWER,
1469 	    &power_req, &result)) == DDI_SUCCESS) {
1470 		/*
1471 		 * Now do involuntary pm accounting;  If we've just cycled power
1472 		 * on a voluntarily pm'd node, and by inference on its entire
1473 		 * subtree, we need to set the subtree (including those nodes
1474 		 * already detached) volpmd counts to 0, and subtract out the
1475 		 * value of the current node's volpmd count from the ancestors
1476 		 */
1477 		if (clearvolpmd) {
1478 			int volpmd = DEVI(dip)->devi_pm_volpmd;
1479 			pm_clear_volpm_dip(dip);
1480 			pm_clear_volpm_list(dip);
1481 			if (volpmd) {
1482 				(void) ddi_pathname(dip, pathbuf);
1483 				(void) pm_noinvol_update(PM_BP_NOINVOL_POWER,
1484 				    volpmd, 0, pathbuf, dip);
1485 			}
1486 		}
1487 	} else {
1488 		PMD(PMD_FAIL, ("%s: can't set comp %d (%s) of %s@%s(%s#%d) "
1489 		    "to level %d (%s)\n", pmf, comp, cp->pmc_comp.pmc_name,
1490 		    PM_DEVICE(dip), level, power_val_to_string(cp, level)))
1491 	}
1492 	/*
1493 	 * If some other devices were also powered up (e.g. other cpus in
1494 	 * the same domain) return a pointer to that list
1495 	 */
1496 	if (devlist) {
1497 		*devlist = (pm_ppm_devlist_t *)
1498 		    power_req.req.ppm_set_power_req.cookie;
1499 	}
1500 	/*
1501 	 * We will have to resume the device if the device is backwards compat
1502 	 * device and either of the following is true:
1503 	 * -This is comp 0 and we have successfully powered it up
1504 	 * -This is comp 0 and we have failed to power it down. Resume is
1505 	 *  needed because we have suspended it above
1506 	 */
1507 
1508 	if (bc && comp == 0) {
1509 		ASSERT(PM_ISDIRECT(dip) || DEVI_IS_DETACHING(dip));
1510 		if (power_op_ret == DDI_SUCCESS) {
1511 			if (POWERING_ON(old_level, level)) {
1512 				/*
1513 				 * It must be either suspended or resumed
1514 				 * via pm_power_has_changed path
1515 				 */
1516 				ASSERT((DEVI(dip)->devi_pm_flags &
1517 				    PMC_SUSPENDED) ||
1518 				    (PM_CP(dip, comp)->pmc_flags &
1519 				    PM_PHC_WHILE_SET_POWER));
1520 
1521 					resume_needed = suspended;
1522 			}
1523 		} else {
1524 			if (POWERING_OFF(old_level, level)) {
1525 				/*
1526 				 * It must be either suspended or resumed
1527 				 * via pm_power_has_changed path
1528 				 */
1529 				ASSERT((DEVI(dip)->devi_pm_flags &
1530 				    PMC_SUSPENDED) ||
1531 				    (PM_CP(dip, comp)->pmc_flags &
1532 				    PM_PHC_WHILE_SET_POWER));
1533 
1534 					resume_needed = suspended;
1535 			}
1536 		}
1537 	}
1538 	if (resume_needed) {
1539 		ASSERT(DEVI(dip)->devi_pm_flags & PMC_SUSPENDED);
1540 		/* ppm is not interested in DDI_PM_RESUME */
1541 		if ((power_op_ret = devi_attach(dip, DDI_PM_RESUME)) ==
1542 		    DDI_SUCCESS) {
1543 			DEVI(dip)->devi_pm_flags &= ~PMC_SUSPENDED;
1544 		} else
1545 			cmn_err(CE_WARN, "!pm: Can't resume %s@%s(%s#%d)",
1546 			    PM_DEVICE(dip));
1547 	}
1548 	return (power_op_ret);
1549 }
1550 
1551 /*
1552  * Return true if we are the owner or a borrower of the devi lock.  See
1553  * pm_lock_power_single() about borrowing the lock.
1554  */
1555 static int
1556 pm_devi_lock_held(dev_info_t *dip)
1557 {
1558 	lock_loan_t *cur;
1559 
1560 	if (DEVI_BUSY_OWNED(dip))
1561 	    return (1);
1562 
1563 	/* return false if no locks borrowed */
1564 	if (lock_loan_head.pmlk_next == NULL)
1565 		return (0);
1566 
1567 	mutex_enter(&pm_loan_lock);
1568 	/* see if our thread is registered as a lock borrower. */
1569 	for (cur = lock_loan_head.pmlk_next; cur; cur = cur->pmlk_next)
1570 		if (cur->pmlk_borrower == curthread)
1571 			break;
1572 	mutex_exit(&pm_loan_lock);
1573 
1574 	return (cur != NULL && cur->pmlk_lender == DEVI(dip)->devi_busy_thread);
1575 }
1576 
1577 /*
1578  * pm_set_power: adjusts power level of device.	 Assumes device is power
1579  * manageable & component exists.
1580  *
1581  * Cases which require us to bring up devices we keep up ("wekeepups") for
1582  * backwards compatible devices:
1583  *	component 0 is off and we're bringing it up from 0
1584  *		bring up wekeepup first
1585  *	and recursively when component 0 is off and we bring some other
1586  *	component up from 0
1587  * For devices which are not backward compatible, our dependency notion is much
1588  * simpler.  Unless all components are off, then wekeeps must be on.
1589  * We don't treat component 0 differently.
1590  * Canblock tells how to deal with a direct pm'd device.
1591  * Scan arg tells us if we were called from scan, in which case we don't need
1592  * to go back to the root node and walk down to change power.
1593  */
1594 int
1595 pm_set_power(dev_info_t *dip, int comp, int level, int direction,
1596     pm_canblock_t canblock, int scan, int *retp)
1597 {
1598 	PMD_FUNC(pmf, "set_power")
1599 	char		*pathbuf;
1600 	pm_bp_child_pwrchg_t bpc;
1601 	pm_sp_misc_t	pspm;
1602 	int		ret = DDI_SUCCESS;
1603 	int		unused = DDI_SUCCESS;
1604 	dev_info_t	*pdip = ddi_get_parent(dip);
1605 
1606 #ifdef DEBUG
1607 	int		diverted = 0;
1608 
1609 	/*
1610 	 * This prevents operations on the console from calling prom_printf and
1611 	 * either deadlocking or bringing up the console because of debug
1612 	 * output
1613 	 */
1614 	if (dip == cfb_dip) {
1615 		diverted++;
1616 		mutex_enter(&pm_debug_lock);
1617 		pm_divertdebug++;
1618 		mutex_exit(&pm_debug_lock);
1619 	}
1620 #endif
1621 	ASSERT(direction == PM_LEVEL_UPONLY || direction == PM_LEVEL_DOWNONLY ||
1622 	    direction == PM_LEVEL_EXACT);
1623 	PMD(PMD_SET, ("%s: %s@%s(%s#%d), comp=%d, dir=%s, new=%d\n",
1624 	    pmf, PM_DEVICE(dip), comp, pm_decode_direction(direction), level))
1625 	pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1626 	(void) ddi_pathname(dip, pathbuf);
1627 	bpc.bpc_dip = dip;
1628 	bpc.bpc_path = pathbuf;
1629 	bpc.bpc_comp = comp;
1630 	bpc.bpc_olevel = PM_CURPOWER(dip, comp);
1631 	bpc.bpc_nlevel = level;
1632 	pspm.pspm_direction = direction;
1633 	pspm.pspm_errnop = retp;
1634 	pspm.pspm_canblock = canblock;
1635 	pspm.pspm_scan = scan;
1636 	bpc.bpc_private = &pspm;
1637 
1638 	/*
1639 	 * If a config operation is being done (we've locked the parent) or
1640 	 * we already hold the power lock (we've locked the node)
1641 	 * then we can operate directly on the node because we have already
1642 	 * brought up all the ancestors, otherwise, we have to go back to the
1643 	 * top of the tree.
1644 	 */
1645 	if (pm_devi_lock_held(pdip) || pm_devi_lock_held(dip))
1646 		ret = pm_busop_set_power(dip, NULL, BUS_POWER_CHILD_PWRCHG,
1647 		    (void *)&bpc, (void *)&unused);
1648 	else
1649 		ret = pm_busop_bus_power(ddi_root_node(), NULL,
1650 		    BUS_POWER_CHILD_PWRCHG, (void *)&bpc, (void *)&unused);
1651 #ifdef DEBUG
1652 	if (ret != DDI_SUCCESS || *retp != DDI_SUCCESS) {
1653 		PMD(PMD_ERROR, ("%s: %s@%s(%s#%d) can't change power, ret=%d, "
1654 		    "errno=%d\n", pmf, PM_DEVICE(dip), ret, *retp))
1655 	}
1656 	if (diverted) {
1657 		mutex_enter(&pm_debug_lock);
1658 		pm_divertdebug--;
1659 		mutex_exit(&pm_debug_lock);
1660 	}
1661 #endif
1662 	kmem_free(pathbuf, MAXPATHLEN);
1663 	return (ret);
1664 }
1665 
1666 /*
1667  * If holddip is set, then if a dip is found we return with the node held.
1668  *
1669  * This code uses the same locking scheme as e_ddi_hold_devi_by_path
1670  * (resolve_pathname), but it does not drive attach.
1671  */
1672 dev_info_t *
1673 pm_name_to_dip(char *pathname, int holddip)
1674 {
1675 	struct pathname pn;
1676 	char		*component;
1677 	dev_info_t	*parent, *child;
1678 	int		circ;
1679 
1680 	if ((pathname == NULL) || (*pathname != '/'))
1681 		return (NULL);
1682 
1683 	/* setup pathname and allocate component */
1684 	if (pn_get(pathname, UIO_SYSSPACE, &pn))
1685 		return (NULL);
1686 	component = kmem_alloc(MAXNAMELEN, KM_SLEEP);
1687 
1688 	/* start at top, process '/' component */
1689 	parent = child = ddi_root_node();
1690 	ndi_hold_devi(parent);
1691 	pn_skipslash(&pn);
1692 	ASSERT(i_ddi_devi_attached(parent));
1693 
1694 	/* process components of pathname */
1695 	while (pn_pathleft(&pn)) {
1696 		(void) pn_getcomponent(&pn, component);
1697 
1698 		/* enter parent and search for component child */
1699 		ndi_devi_enter(parent, &circ);
1700 		child = ndi_devi_findchild(parent, component);
1701 		if ((child == NULL) || !i_ddi_devi_attached(child)) {
1702 			child = NULL;
1703 			ndi_devi_exit(parent, circ);
1704 			ndi_rele_devi(parent);
1705 			goto out;
1706 		}
1707 
1708 		/* attached child found, hold child and release parent */
1709 		ndi_hold_devi(child);
1710 		ndi_devi_exit(parent, circ);
1711 		ndi_rele_devi(parent);
1712 
1713 		/* child becomes parent, and process next component */
1714 		parent = child;
1715 		pn_skipslash(&pn);
1716 
1717 		/* loop with active ndi_devi_hold of child->parent */
1718 	}
1719 
1720 out:
1721 	pn_free(&pn);
1722 	kmem_free(component, MAXNAMELEN);
1723 
1724 	/* if we are not asked to return with hold, drop current hold */
1725 	if (child && !holddip)
1726 		ndi_rele_devi(child);
1727 	return (child);
1728 }
1729 
1730 /*
1731  * Search for a dependency and mark it unsatisfied
1732  */
1733 static void
1734 pm_unsatisfy(char *keeper, char *kept)
1735 {
1736 	PMD_FUNC(pmf, "unsatisfy")
1737 	pm_pdr_t *dp;
1738 
1739 	PMD(PMD_KEEPS, ("%s: keeper=%s, kept=%s\n", pmf, keeper, kept))
1740 	for (dp = pm_dep_head; dp; dp = dp->pdr_next) {
1741 		if (!dp->pdr_isprop) {
1742 			if (strcmp(dp->pdr_keeper, keeper) == 0 &&
1743 			    (dp->pdr_kept_count > 0) &&
1744 			    strcmp(dp->pdr_kept_paths[0], kept) == 0) {
1745 				if (dp->pdr_satisfied) {
1746 					dp->pdr_satisfied = 0;
1747 					pm_unresolved_deps++;
1748 					PMD(PMD_KEEPS, ("%s: clear satisfied, "
1749 					    "pm_unresolved_deps now %d\n", pmf,
1750 					    pm_unresolved_deps))
1751 				}
1752 			}
1753 		}
1754 	}
1755 }
1756 
1757 /*
1758  * Device dip is being un power managed, it keeps up count other devices.
1759  * We need to release any hold we have on the kept devices, and also
1760  * mark the dependency no longer satisfied.
1761  */
1762 static void
1763 pm_unkeeps(int count, char *keeper, char **keptpaths, int pwr)
1764 {
1765 	PMD_FUNC(pmf, "unkeeps")
1766 	int i, j;
1767 	dev_info_t *kept;
1768 	dev_info_t *dip;
1769 	struct pm_component *cp;
1770 	int keeper_on = 0, circ;
1771 
1772 	PMD(PMD_KEEPS, ("%s: count=%d, keeper=%s, keptpaths=%p\n", pmf, count,
1773 	    keeper, (void *)keptpaths))
1774 	/*
1775 	 * Try to grab keeper. Keeper may have gone away by now,
1776 	 * in this case, used the passed in value pwr
1777 	 */
1778 	dip = pm_name_to_dip(keeper, 1);
1779 	for (i = 0; i < count; i++) {
1780 		/* Release power hold */
1781 		kept = pm_name_to_dip(keptpaths[i], 1);
1782 		if (kept) {
1783 			PMD(PMD_KEEPS, ("%s: %s@%s(%s#%d)[%d]\n", pmf,
1784 			    PM_DEVICE(kept), i))
1785 			/*
1786 			 * We need to check if we skipped a bringup here
1787 			 * because we could have failed the bringup
1788 			 * (ie DIRECT PM device) and have
1789 			 * not increment the count.
1790 			 */
1791 			if ((dip != NULL) && (PM_GET_PM_INFO(dip) != NULL)) {
1792 				keeper_on = 0;
1793 				PM_LOCK_POWER(dip, &circ);
1794 				for (j = 0; j < PM_NUMCMPTS(dip); j++) {
1795 				    cp = &DEVI(dip)->devi_pm_components[j];
1796 					if (cur_power(cp)) {
1797 						keeper_on++;
1798 						break;
1799 					}
1800 				}
1801 				if (keeper_on && (PM_SKBU(kept) == 0)) {
1802 					pm_rele_power(kept);
1803 					DEVI(kept)->devi_pm_flags
1804 						&= ~PMC_SKIP_BRINGUP;
1805 				}
1806 				PM_UNLOCK_POWER(dip, circ);
1807 			} else if (pwr) {
1808 				if (PM_SKBU(kept) == 0) {
1809 					pm_rele_power(kept);
1810 					DEVI(kept)->devi_pm_flags
1811 					    &= ~PMC_SKIP_BRINGUP;
1812 				}
1813 			}
1814 			ddi_release_devi(kept);
1815 		}
1816 		/*
1817 		 * mark this dependency not satisfied
1818 		 */
1819 		pm_unsatisfy(keeper, keptpaths[i]);
1820 	}
1821 	if (dip)
1822 		ddi_release_devi(dip);
1823 }
1824 
1825 /*
1826  * Device kept is being un power managed, it is kept up by keeper.
1827  * We need to mark the dependency no longer satisfied.
1828  */
1829 static void
1830 pm_unkepts(char *kept, char *keeper)
1831 {
1832 	PMD_FUNC(pmf, "unkepts")
1833 	PMD(PMD_KEEPS, ("%s: kept=%s, keeper=%s\n", pmf, kept, keeper))
1834 	ASSERT(keeper != NULL);
1835 	/*
1836 	 * mark this dependency not satisfied
1837 	 */
1838 	pm_unsatisfy(keeper, kept);
1839 }
1840 
1841 /*
1842  * Removes dependency information and hold on the kepts, if the path is a
1843  * path of a keeper.
1844  */
1845 static void
1846 pm_free_keeper(char *path, int pwr)
1847 {
1848 	pm_pdr_t *dp;
1849 	int i;
1850 	size_t length;
1851 
1852 	for (dp = pm_dep_head; dp; dp = dp->pdr_next) {
1853 		if (strcmp(dp->pdr_keeper, path) != 0)
1854 			continue;
1855 		/*
1856 		 * Remove all our kept holds and the dependency records,
1857 		 * then free up the kept lists.
1858 		 */
1859 		pm_unkeeps(dp->pdr_kept_count, path, dp->pdr_kept_paths, pwr);
1860 		if (dp->pdr_kept_count)  {
1861 			for (i = 0; i < dp->pdr_kept_count; i++) {
1862 				length = strlen(dp->pdr_kept_paths[i]);
1863 				kmem_free(dp->pdr_kept_paths[i], length + 1);
1864 			}
1865 			kmem_free(dp->pdr_kept_paths,
1866 			    dp->pdr_kept_count * sizeof (char **));
1867 			dp->pdr_kept_paths = NULL;
1868 			dp->pdr_kept_count = 0;
1869 		}
1870 	}
1871 }
1872 
1873 /*
1874  * Removes the device represented by path from the list of kepts, if the
1875  * path is a path of a kept
1876  */
1877 static void
1878 pm_free_kept(char *path)
1879 {
1880 	pm_pdr_t *dp;
1881 	int i;
1882 	int j, count;
1883 	size_t length;
1884 	char **paths;
1885 
1886 	for (dp = pm_dep_head; dp; dp = dp->pdr_next) {
1887 		if (dp->pdr_kept_count == 0)
1888 			continue;
1889 		count = dp->pdr_kept_count;
1890 		/* Remove this device from the kept path lists */
1891 		for (i = 0; i < count; i++) {
1892 			if (strcmp(dp->pdr_kept_paths[i], path) == 0) {
1893 				pm_unkepts(path, dp->pdr_keeper);
1894 				length = strlen(dp->pdr_kept_paths[i]) + 1;
1895 				kmem_free(dp->pdr_kept_paths[i], length);
1896 				dp->pdr_kept_paths[i] = NULL;
1897 				dp->pdr_kept_count--;
1898 			}
1899 		}
1900 		/* Compact the kept paths array */
1901 		if (dp->pdr_kept_count) {
1902 			length = dp->pdr_kept_count * sizeof (char **);
1903 			paths = kmem_zalloc(length, KM_SLEEP);
1904 			j = 0;
1905 			for (i = 0; i < count; i++) {
1906 				if (dp->pdr_kept_paths[i] != NULL) {
1907 					paths[j] = dp->pdr_kept_paths[i];
1908 					j++;
1909 				}
1910 			}
1911 			ASSERT(j == dp->pdr_kept_count);
1912 		}
1913 		/* Now free the old array and point to the new one */
1914 		kmem_free(dp->pdr_kept_paths, count * sizeof (char **));
1915 		if (dp->pdr_kept_count)
1916 			dp->pdr_kept_paths = paths;
1917 		else
1918 			dp->pdr_kept_paths = NULL;
1919 	}
1920 }
1921 
1922 /*
1923  * Free the dependency information for a device.
1924  */
1925 void
1926 pm_free_keeps(char *path, int pwr)
1927 {
1928 	PMD_FUNC(pmf, "free_keeps")
1929 
1930 #ifdef DEBUG
1931 	int doprdeps = 0;
1932 	void prdeps(char *);
1933 
1934 	PMD(PMD_KEEPS, ("%s: %s\n", pmf, path))
1935 	if (pm_debug & PMD_KEEPS) {
1936 		doprdeps = 1;
1937 		prdeps("pm_free_keeps before");
1938 	}
1939 #endif
1940 	/*
1941 	 * First assume we are a keeper and remove all our kepts.
1942 	 */
1943 	pm_free_keeper(path, pwr);
1944 	/*
1945 	 * Now assume we a kept device, and remove all our records.
1946 	 */
1947 	pm_free_kept(path);
1948 #ifdef	DEBUG
1949 	if (doprdeps) {
1950 		prdeps("pm_free_keeps after");
1951 	}
1952 #endif
1953 }
1954 
1955 static int
1956 pm_is_kept(char *path)
1957 {
1958 	pm_pdr_t *dp;
1959 	int i;
1960 
1961 	for (dp = pm_dep_head; dp; dp = dp->pdr_next) {
1962 		if (dp->pdr_kept_count == 0)
1963 			continue;
1964 		for (i = 0; i < dp->pdr_kept_count; i++) {
1965 			if (strcmp(dp->pdr_kept_paths[i], path) == 0)
1966 				return (1);
1967 		}
1968 	}
1969 	return (0);
1970 }
1971 
1972 static void
1973 e_pm_hold_rele_power(dev_info_t *dip, int cnt)
1974 {
1975 	PMD_FUNC(pmf, "hold_rele_power")
1976 	int circ;
1977 
1978 	if ((dip == NULL) ||
1979 	    (PM_GET_PM_INFO(dip) == NULL) || PM_ISBC(dip))
1980 		return;
1981 
1982 	PM_LOCK_POWER(dip, &circ);
1983 	ASSERT(cnt >= 0 && PM_KUC(dip) >= 0 || cnt < 0 && PM_KUC(dip) > 0);
1984 	PMD(PMD_KIDSUP, ("%s: kidsupcnt for %s@%s(%s#%d) %d->%d\n", pmf,
1985 	    PM_DEVICE(dip), PM_KUC(dip), (PM_KUC(dip) + cnt)))
1986 
1987 	PM_KUC(dip) += cnt;
1988 
1989 	ASSERT(PM_KUC(dip) >= 0);
1990 	PM_UNLOCK_POWER(dip, circ);
1991 
1992 	if (cnt < 0 && PM_KUC(dip) == 0)
1993 		pm_rescan(dip);
1994 }
1995 
1996 #define	MAX_PPM_HANDLERS	4
1997 
1998 kmutex_t ppm_lock;	/* in case we ever do multi-threaded startup */
1999 
2000 struct	ppm_callbacks {
2001 	int (*ppmc_func)(dev_info_t *);
2002 	dev_info_t	*ppmc_dip;
2003 } ppm_callbacks[MAX_PPM_HANDLERS + 1];
2004 
2005 
2006 /*
2007  * This routine calls into all the registered ppms to notify them
2008  * that either all components of power-managed devices are at their
2009  * lowest levels or no longer all are at their lowest levels.
2010  */
2011 static void
2012 pm_ppm_notify_all_lowest(dev_info_t *dip, int mode)
2013 {
2014 	struct ppm_callbacks *ppmcp;
2015 	power_req_t power_req;
2016 	int result = 0;
2017 
2018 	power_req.request_type = PMR_PPM_ALL_LOWEST;
2019 	power_req.req.ppm_all_lowest_req.mode = mode;
2020 	mutex_enter(&ppm_lock);
2021 	for (ppmcp = ppm_callbacks; ppmcp->ppmc_func; ppmcp++)
2022 		(void) pm_ctlops((dev_info_t *)ppmcp->ppmc_dip, dip,
2023 		    DDI_CTLOPS_POWER, &power_req, &result);
2024 	mutex_exit(&ppm_lock);
2025 }
2026 
2027 static void
2028 pm_set_pm_info(dev_info_t *dip, void *value)
2029 {
2030 	DEVI(dip)->devi_pm_info = value;
2031 }
2032 
2033 pm_rsvp_t *pm_blocked_list;
2034 
2035 /*
2036  * Look up an entry in the blocked list by dip and component
2037  */
2038 static pm_rsvp_t *
2039 pm_rsvp_lookup(dev_info_t *dip, int comp)
2040 {
2041 	pm_rsvp_t *p;
2042 	ASSERT(MUTEX_HELD(&pm_rsvp_lock));
2043 	for (p = pm_blocked_list; p; p = p->pr_next)
2044 		if (p->pr_dip == dip && p->pr_comp == comp) {
2045 			return (p);
2046 		}
2047 	return (NULL);
2048 }
2049 
2050 /*
2051  * Called when a device which is direct power managed (or the parent or
2052  * dependent of such a device) changes power, or when a pm clone is closed
2053  * that was direct power managing a device.  This call results in pm_blocked()
2054  * (below) returning.
2055  */
2056 void
2057 pm_proceed(dev_info_t *dip, int cmd, int comp, int newlevel)
2058 {
2059 	PMD_FUNC(pmf, "proceed")
2060 	pm_rsvp_t *found = NULL;
2061 	pm_rsvp_t *p;
2062 
2063 	mutex_enter(&pm_rsvp_lock);
2064 	switch (cmd) {
2065 	/*
2066 	 * we're giving up control, let any pending op continue
2067 	 */
2068 	case PMP_RELEASE:
2069 		for (p = pm_blocked_list; p; p = p->pr_next) {
2070 			if (dip == p->pr_dip) {
2071 				p->pr_retval = PMP_RELEASE;
2072 				PMD(PMD_DPM, ("%s: RELEASE %s@%s(%s#%d)\n",
2073 				    pmf, PM_DEVICE(dip)))
2074 				cv_signal(&p->pr_cv);
2075 			}
2076 		}
2077 		break;
2078 
2079 	/*
2080 	 * process has done PM_SET_CURRENT_POWER; let a matching request
2081 	 * succeed and a non-matching request for the same device fail
2082 	 */
2083 	case PMP_SETPOWER:
2084 		found = pm_rsvp_lookup(dip, comp);
2085 		if (!found)	/* if driver not waiting */
2086 			break;
2087 		/*
2088 		 * This cannot be pm_lower_power, since that can only happen
2089 		 * during detach or probe
2090 		 */
2091 		if (found->pr_newlevel <= newlevel) {
2092 			found->pr_retval = PMP_SUCCEED;
2093 			PMD(PMD_DPM, ("%s: SUCCEED %s@%s(%s#%d)\n", pmf,
2094 			    PM_DEVICE(dip)))
2095 		} else {
2096 			found->pr_retval = PMP_FAIL;
2097 			PMD(PMD_DPM, ("%s: FAIL %s@%s(%s#%d)\n", pmf,
2098 			    PM_DEVICE(dip)))
2099 		}
2100 		cv_signal(&found->pr_cv);
2101 		break;
2102 
2103 	default:
2104 		panic("pm_proceed unknown cmd %d", cmd);
2105 	}
2106 	mutex_exit(&pm_rsvp_lock);
2107 }
2108 
2109 /*
2110  * This routine dispatches new work to the dependency thread. Caller must
2111  * be prepared to block for memory if necessary.
2112  */
2113 void
2114 pm_dispatch_to_dep_thread(int cmd, char *keeper, char *kept, int wait,
2115     int *res, int cached_pwr)
2116 {
2117 	pm_dep_wk_t	*new_work;
2118 
2119 	new_work = kmem_zalloc(sizeof (pm_dep_wk_t), KM_SLEEP);
2120 	new_work->pdw_type = cmd;
2121 	new_work->pdw_wait = wait;
2122 	new_work->pdw_done = 0;
2123 	new_work->pdw_ret = 0;
2124 	new_work->pdw_pwr = cached_pwr;
2125 	cv_init(&new_work->pdw_cv, NULL, CV_DEFAULT, NULL);
2126 	if (keeper != NULL) {
2127 		new_work->pdw_keeper = kmem_zalloc(strlen(keeper) + 1,
2128 		    KM_SLEEP);
2129 		(void) strcpy(new_work->pdw_keeper, keeper);
2130 	}
2131 	if (kept != NULL) {
2132 		new_work->pdw_kept = kmem_zalloc(strlen(kept) + 1, KM_SLEEP);
2133 		(void) strcpy(new_work->pdw_kept, kept);
2134 	}
2135 	mutex_enter(&pm_dep_thread_lock);
2136 	if (pm_dep_thread_workq == NULL) {
2137 		pm_dep_thread_workq = new_work;
2138 		pm_dep_thread_tail = new_work;
2139 		new_work->pdw_next = NULL;
2140 	} else {
2141 		pm_dep_thread_tail->pdw_next = new_work;
2142 		pm_dep_thread_tail = new_work;
2143 		new_work->pdw_next = NULL;
2144 	}
2145 	cv_signal(&pm_dep_thread_cv);
2146 	/* If caller asked for it, wait till it is done. */
2147 	if (wait)  {
2148 		while (!new_work->pdw_done)
2149 			cv_wait(&new_work->pdw_cv, &pm_dep_thread_lock);
2150 		/*
2151 		 * Pass return status, if any, back.
2152 		 */
2153 		if (res != NULL)
2154 			*res = new_work->pdw_ret;
2155 		/*
2156 		 * If we asked to wait, it is our job to free the request
2157 		 * structure.
2158 		 */
2159 		if (new_work->pdw_keeper)
2160 			kmem_free(new_work->pdw_keeper,
2161 			    strlen(new_work->pdw_keeper) + 1);
2162 		if (new_work->pdw_kept)
2163 			kmem_free(new_work->pdw_kept,
2164 			    strlen(new_work->pdw_kept) + 1);
2165 		kmem_free(new_work, sizeof (pm_dep_wk_t));
2166 	}
2167 	mutex_exit(&pm_dep_thread_lock);
2168 }
2169 
2170 /*
2171  * Release the pm resource for this device.
2172  */
2173 void
2174 pm_rem_info(dev_info_t *dip)
2175 {
2176 	PMD_FUNC(pmf, "rem_info")
2177 	int		i, count = 0;
2178 	pm_info_t	*info = PM_GET_PM_INFO(dip);
2179 	dev_info_t	*pdip = ddi_get_parent(dip);
2180 	char		*pathbuf;
2181 	int		work_type = PM_DEP_WK_DETACH;
2182 
2183 	ASSERT(info);
2184 
2185 	ASSERT(!PM_IAM_LOCKING_DIP(dip));
2186 	if (PM_ISDIRECT(dip)) {
2187 		info->pmi_dev_pm_state &= ~PM_DIRECT;
2188 		ASSERT(info->pmi_clone);
2189 		info->pmi_clone = 0;
2190 		pm_proceed(dip, PMP_RELEASE, -1, -1);
2191 	}
2192 	ASSERT(!PM_GET_PM_SCAN(dip));
2193 
2194 	/*
2195 	 * Now adjust parent's kidsupcnt.  BC nodes we check only comp 0,
2196 	 * Others we check all components.  BC node that has already
2197 	 * called pm_destroy_components() has zero component count.
2198 	 * Parents that get notification are not adjusted because their
2199 	 * kidsupcnt is always 0 (or 1 during configuration).
2200 	 */
2201 	PMD(PMD_KEEPS, ("%s: %s@%s(%s#%d) has %d components\n", pmf,
2202 	    PM_DEVICE(dip), PM_NUMCMPTS(dip)))
2203 
2204 	/* node is detached, so we can examine power without locking */
2205 	if (PM_ISBC(dip)) {
2206 		count = (PM_CURPOWER(dip, 0) != 0);
2207 	} else {
2208 		for (i = 0; i < PM_NUMCMPTS(dip); i++)
2209 			count += (PM_CURPOWER(dip, i) != 0);
2210 	}
2211 
2212 	if (PM_NUMCMPTS(dip) && pdip && !PM_WANTS_NOTIFICATION(pdip))
2213 		e_pm_hold_rele_power(pdip, -count);
2214 
2215 	/* Schedule a request to clean up dependency records */
2216 	pathbuf = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
2217 	(void) ddi_pathname(dip, pathbuf);
2218 	pm_dispatch_to_dep_thread(work_type, pathbuf, pathbuf,
2219 	    PM_DEP_NOWAIT, NULL, (count > 0));
2220 	kmem_free(pathbuf, MAXPATHLEN);
2221 
2222 	/*
2223 	 * Adjust the pm_comps_notlowest count since this device is
2224 	 * not being power-managed anymore.
2225 	 */
2226 	for (i = 0; i < PM_NUMCMPTS(dip); i++) {
2227 		if (PM_CURPOWER(dip, i) != 0)
2228 			PM_DECR_NOTLOWEST(dip);
2229 	}
2230 	/*
2231 	 * Once we clear the info pointer, it looks like it is not power
2232 	 * managed to everybody else.
2233 	 */
2234 	pm_set_pm_info(dip, NULL);
2235 	kmem_free(info, sizeof (pm_info_t));
2236 }
2237 
2238 int
2239 pm_get_norm_pwrs(dev_info_t *dip, int **valuep, size_t *length)
2240 {
2241 	int components = PM_NUMCMPTS(dip);
2242 	int *bufp;
2243 	size_t size;
2244 	int i;
2245 
2246 	if (components <= 0) {
2247 		cmn_err(CE_NOTE, "!pm: %s@%s(%s#%d) has no components, "
2248 		    "can't get normal power values\n", PM_DEVICE(dip));
2249 		return (DDI_FAILURE);
2250 	} else {
2251 		size = components * sizeof (int);
2252 		bufp = kmem_alloc(size, KM_SLEEP);
2253 		for (i = 0; i < components; i++) {
2254 			bufp[i] = pm_get_normal_power(dip, i);
2255 		}
2256 	}
2257 	*length = size;
2258 	*valuep = bufp;
2259 	return (DDI_SUCCESS);
2260 }
2261 
2262 static int
2263 pm_reset_timestamps(dev_info_t *dip, void *arg)
2264 {
2265 	_NOTE(ARGUNUSED(arg))
2266 
2267 	int components;
2268 	int	i;
2269 
2270 	if (!PM_GET_PM_INFO(dip))
2271 		return (DDI_WALK_CONTINUE);
2272 	components = PM_NUMCMPTS(dip);
2273 	ASSERT(components > 0);
2274 	PM_LOCK_BUSY(dip);
2275 	for (i = 0; i < components; i++) {
2276 		struct pm_component *cp;
2277 		/*
2278 		 * If the component was not marked as busy,
2279 		 * reset its timestamp to now.
2280 		 */
2281 		cp = PM_CP(dip, i);
2282 		if (cp->pmc_timestamp)
2283 			cp->pmc_timestamp = gethrestime_sec();
2284 	}
2285 	PM_UNLOCK_BUSY(dip);
2286 	return (DDI_WALK_CONTINUE);
2287 }
2288 
2289 /*
2290  * Convert a power level to an index into the levels array (or
2291  * just PM_LEVEL_UNKNOWN in that special case).
2292  */
2293 static int
2294 pm_level_to_index(dev_info_t *dip, pm_component_t *cp, int level)
2295 {
2296 	PMD_FUNC(pmf, "level_to_index")
2297 	int i;
2298 	int limit = cp->pmc_comp.pmc_numlevels;
2299 	int *ip = cp->pmc_comp.pmc_lvals;
2300 
2301 	if (level == PM_LEVEL_UNKNOWN)
2302 		return (level);
2303 
2304 	for (i = 0; i < limit; i++) {
2305 		if (level == *ip++) {
2306 			PMD(PMD_LEVEL, ("%s: %s@%s(%s#%d)[%d] to %x\n",
2307 			    pmf, PM_DEVICE(dip),
2308 			    (int)(cp - DEVI(dip)->devi_pm_components), level))
2309 			return (i);
2310 		}
2311 	}
2312 	panic("pm_level_to_index: level %d not found for device "
2313 	    "%s@%s(%s#%d)", level, PM_DEVICE(dip));
2314 	/*NOTREACHED*/
2315 }
2316 
2317 /*
2318  * Internal function to set current power level
2319  */
2320 static void
2321 e_pm_set_cur_pwr(dev_info_t *dip, pm_component_t *cp, int level)
2322 {
2323 	PMD_FUNC(pmf, "set_cur_pwr")
2324 	int curpwr = (cp->pmc_flags & PM_PHC_WHILE_SET_POWER ?
2325 	    cp->pmc_phc_pwr : cp->pmc_cur_pwr);
2326 
2327 	/*
2328 	 * Nothing to adjust if current & new levels are the same.
2329 	 */
2330 	if (curpwr != PM_LEVEL_UNKNOWN &&
2331 	    level == cp->pmc_comp.pmc_lvals[curpwr])
2332 		return;
2333 
2334 	/*
2335 	 * Keep the count for comps doing transition to/from lowest
2336 	 * level.
2337 	 */
2338 	if (curpwr == 0) {
2339 		PM_INCR_NOTLOWEST(dip);
2340 	} else if (level == cp->pmc_comp.pmc_lvals[0]) {
2341 		PM_DECR_NOTLOWEST(dip);
2342 	}
2343 	cp->pmc_phc_pwr = PM_LEVEL_UNKNOWN;
2344 	cp->pmc_cur_pwr = pm_level_to_index(dip, cp, level);
2345 }
2346 
2347 /*
2348  * This is the default method of setting the power of a device if no ppm
2349  * driver has claimed it.
2350  */
2351 int
2352 pm_power(dev_info_t *dip, int comp, int level)
2353 {
2354 	PMD_FUNC(pmf, "power")
2355 	struct dev_ops	*ops;
2356 	int		(*fn)(dev_info_t *, int, int);
2357 	struct pm_component *cp = PM_CP(dip, comp);
2358 	int retval;
2359 	pm_info_t *info = PM_GET_PM_INFO(dip);
2360 	static int pm_phc_impl(dev_info_t *, int, int, int);
2361 
2362 	PMD(PMD_KIDSUP, ("%s: %s@%s(%s#%d), comp=%d, level=%d\n", pmf,
2363 	    PM_DEVICE(dip), comp, level))
2364 	if (!(ops = ddi_get_driver(dip))) {
2365 		PMD(PMD_FAIL, ("%s: %s@%s(%s#%d) has no ops\n", pmf,
2366 		    PM_DEVICE(dip)))
2367 		return (DDI_FAILURE);
2368 	}
2369 	if ((ops->devo_rev < 2) || !(fn = ops->devo_power)) {
2370 		PMD(PMD_FAIL, ("%s: %s%s\n", pmf,
2371 		    (ops->devo_rev < 2 ? " wrong devo_rev" : ""),
2372 		    (!fn ? " devo_power NULL" : "")))
2373 		return (DDI_FAILURE);
2374 	}
2375 	cp->pmc_flags |= PM_POWER_OP;
2376 	retval = (*fn)(dip, comp, level);
2377 	cp->pmc_flags &= ~PM_POWER_OP;
2378 	if (retval == DDI_SUCCESS) {
2379 		e_pm_set_cur_pwr(dip, PM_CP(dip, comp), level);
2380 		return (DDI_SUCCESS);
2381 	}
2382 
2383 	/*
2384 	 * If pm_power_has_changed() detected a deadlock with pm_power() it
2385 	 * updated only the power level of the component.  If our attempt to
2386 	 * set the device new to a power level above has failed we sync the
2387 	 * total power state via phc code now.
2388 	 */
2389 	if (cp->pmc_flags & PM_PHC_WHILE_SET_POWER) {
2390 		int phc_lvl =
2391 		    cp->pmc_comp.pmc_lvals[cp->pmc_cur_pwr];
2392 
2393 		ASSERT(info);
2394 		(void) pm_phc_impl(dip, comp, phc_lvl, 0);
2395 		PMD(PMD_PHC, ("%s: phc %s@%s(%s#%d) comp=%d level=%d\n",
2396 			pmf, PM_DEVICE(dip), comp, phc_lvl))
2397 	}
2398 
2399 	PMD(PMD_FAIL, ("%s: can't set comp=%d (%s) of %s@%s(%s#%d) to "
2400 	    "level=%d (%s)\n", pmf, comp, cp->pmc_comp.pmc_name, PM_DEVICE(dip),
2401 	    level, power_val_to_string(cp, level)));
2402 	return (DDI_FAILURE);
2403 }
2404 
2405 int
2406 pm_unmanage(dev_info_t *dip)
2407 {
2408 	PMD_FUNC(pmf, "unmanage")
2409 	power_req_t power_req;
2410 	int result, retval = 0;
2411 
2412 	ASSERT(!PM_IAM_LOCKING_DIP(dip));
2413 	PMD(PMD_REMDEV | PMD_KIDSUP, ("%s: %s@%s(%s#%d)\n", pmf,
2414 	    PM_DEVICE(dip)))
2415 	power_req.request_type = PMR_PPM_UNMANAGE;
2416 	power_req.req.ppm_config_req.who = dip;
2417 	if (pm_ppm_claimed(dip))
2418 		retval = pm_ctlops(PPM(dip), dip, DDI_CTLOPS_POWER,
2419 		    &power_req, &result);
2420 #ifdef DEBUG
2421 	else
2422 		retval = pm_ctlops(PPM(dip), dip, DDI_CTLOPS_POWER,
2423 		    &power_req, &result);
2424 #endif
2425 	ASSERT(retval == DDI_SUCCESS);
2426 	pm_rem_info(dip);
2427 	return (retval);
2428 }
2429 
2430 int
2431 pm_raise_power(dev_info_t *dip, int comp, int level)
2432 {
2433 	if (level < 0)
2434 		return (DDI_FAILURE);
2435 	if (!e_pm_valid_info(dip, NULL) || !e_pm_valid_comp(dip, comp, NULL) ||
2436 	    !e_pm_valid_power(dip, comp, level))
2437 		return (DDI_FAILURE);
2438 
2439 	return (dev_is_needed(dip, comp, level, PM_LEVEL_UPONLY));
2440 }
2441 
2442 int
2443 pm_lower_power(dev_info_t *dip, int comp, int level)
2444 {
2445 	PMD_FUNC(pmf, "pm_lower_power")
2446 
2447 	if (!e_pm_valid_info(dip, NULL) || !e_pm_valid_comp(dip, comp, NULL) ||
2448 	    !e_pm_valid_power(dip, comp, level)) {
2449 		PMD(PMD_FAIL, ("%s: validation checks failed for %s@%s(%s#%d) "
2450 		    "comp=%d level=%d\n", pmf, PM_DEVICE(dip), comp, level))
2451 		return (DDI_FAILURE);
2452 	}
2453 
2454 	if (!DEVI_IS_DETACHING(dip)) {
2455 		PMD(PMD_FAIL, ("%s: %s@%s(%s#%d) not detaching\n",
2456 		    pmf, PM_DEVICE(dip)))
2457 		return (DDI_FAILURE);
2458 	}
2459 
2460 	/*
2461 	 * If we don't care about saving power, or we're treating this node
2462 	 * specially, then this is a no-op
2463 	 */
2464 	if (!autopm_enabled || pm_noinvol(dip)) {
2465 		PMD(PMD_FAIL, ("%s: %s@%s(%s#%d) %s%s\n", pmf, PM_DEVICE(dip),
2466 		    !autopm_enabled ? "!autopm_enabled " : "",
2467 		    pm_noinvol(dip) ? "pm_noinvol()" : ""))
2468 		return (DDI_SUCCESS);
2469 	}
2470 
2471 	if (dev_is_needed(dip, comp, level, PM_LEVEL_DOWNONLY) != DDI_SUCCESS) {
2472 		PMD(PMD_FAIL, ("%s: %s@%s(%s#%d) dev_is_needed failed\n", pmf,
2473 		    PM_DEVICE(dip)))
2474 		return (DDI_FAILURE);
2475 	}
2476 	return (DDI_SUCCESS);
2477 }
2478 
2479 /*
2480  * Find the entries struct for a given dip in the blocked list, return it locked
2481  */
2482 static psce_t *
2483 pm_psc_dip_to_direct(dev_info_t *dip, pscc_t **psccp)
2484 {
2485 	pscc_t *p;
2486 	psce_t *psce;
2487 
2488 	rw_enter(&pm_pscc_direct_rwlock, RW_READER);
2489 	for (p = pm_pscc_direct; p; p = p->pscc_next) {
2490 		if (p->pscc_dip == dip) {
2491 			*psccp = p;
2492 			psce = p->pscc_entries;
2493 			mutex_enter(&psce->psce_lock);
2494 			ASSERT(psce);
2495 			rw_exit(&pm_pscc_direct_rwlock);
2496 			return (psce);
2497 		}
2498 	}
2499 	rw_exit(&pm_pscc_direct_rwlock);
2500 	panic("sunpm: no entry for dip %p in direct list", (void *)dip);
2501 	/*NOTREACHED*/
2502 }
2503 
2504 /*
2505  * Write an entry indicating a power level change (to be passed to a process
2506  * later) in the given psce.
2507  * If we were called in the path that brings up the console fb in the
2508  * case of entering the prom, we don't want to sleep.  If the alloc fails, then
2509  * we create a record that has a size of -1, a physaddr of NULL, and that
2510  * has the overflow flag set.
2511  */
2512 static int
2513 psc_entry(ushort_t event, psce_t *psce, dev_info_t *dip, int comp, int new,
2514     int old, int which, pm_canblock_t canblock)
2515 {
2516 	char	buf[MAXNAMELEN];
2517 	pm_state_change_t *p;
2518 	size_t	size;
2519 	caddr_t physpath = NULL;
2520 	int	overrun = 0;
2521 
2522 	ASSERT(MUTEX_HELD(&psce->psce_lock));
2523 	(void) ddi_pathname(dip, buf);
2524 	size = strlen(buf) + 1;
2525 	p = psce->psce_in;
2526 	if (canblock == PM_CANBLOCK_BYPASS) {
2527 		physpath = kmem_alloc(size, KM_NOSLEEP);
2528 		if (physpath == NULL) {
2529 			/*
2530 			 * mark current entry as overrun
2531 			 */
2532 			p->flags |= PSC_EVENT_LOST;
2533 			size = (size_t)-1;
2534 		}
2535 	} else
2536 		physpath = kmem_alloc(size, KM_SLEEP);
2537 	if (p->size) {	/* overflow; mark the next entry */
2538 		if (p->size != (size_t)-1)
2539 			kmem_free(p->physpath, p->size);
2540 		ASSERT(psce->psce_out == p);
2541 		if (p == psce->psce_last) {
2542 			psce->psce_first->flags |= PSC_EVENT_LOST;
2543 			psce->psce_out = psce->psce_first;
2544 		} else {
2545 			(p + 1)->flags |= PSC_EVENT_LOST;
2546 			psce->psce_out = (p + 1);
2547 		}
2548 		overrun++;
2549 	} else if (physpath == NULL) {	/* alloc failed, mark this entry */
2550 		p->flags |= PSC_EVENT_LOST;
2551 		p->size = 0;
2552 		p->physpath = NULL;
2553 	}
2554 	if (which == PSC_INTEREST) {
2555 		mutex_enter(&pm_compcnt_lock);
2556 		if (pm_comps_notlowest == 0)
2557 			p->flags |= PSC_ALL_LOWEST;
2558 		else
2559 			p->flags &= ~PSC_ALL_LOWEST;
2560 		mutex_exit(&pm_compcnt_lock);
2561 	}
2562 	p->event = event;
2563 	p->timestamp = gethrestime_sec();
2564 	p->component = comp;
2565 	p->old_level = old;
2566 	p->new_level = new;
2567 	p->physpath = physpath;
2568 	p->size = size;
2569 	if (physpath != NULL)
2570 		(void) strcpy(p->physpath, buf);
2571 	if (p == psce->psce_last)
2572 		psce->psce_in = psce->psce_first;
2573 	else
2574 		psce->psce_in = ++p;
2575 	mutex_exit(&psce->psce_lock);
2576 	return (overrun);
2577 }
2578 
2579 /*
2580  * Find the next entry on the interest list.  We keep a pointer to the item we
2581  * last returned in the user's cooke.  Returns a locked entries struct.
2582  */
2583 static psce_t *
2584 psc_interest(void **cookie, pscc_t **psccp)
2585 {
2586 	pscc_t *pscc;
2587 	pscc_t **cookiep = (pscc_t **)cookie;
2588 
2589 	if (*cookiep == NULL)
2590 		pscc = pm_pscc_interest;
2591 	else
2592 		pscc = (*cookiep)->pscc_next;
2593 	if (pscc) {
2594 		*cookiep = pscc;
2595 		*psccp = pscc;
2596 		mutex_enter(&pscc->pscc_entries->psce_lock);
2597 		return (pscc->pscc_entries);
2598 	} else {
2599 		return (NULL);
2600 	}
2601 }
2602 
2603 /*
2604  * Create an entry for a process to pick up indicating a power level change.
2605  */
2606 static void
2607 pm_enqueue_notify(ushort_t cmd, dev_info_t *dip, int comp,
2608     int newlevel, int oldlevel, pm_canblock_t canblock)
2609 {
2610 	PMD_FUNC(pmf, "enqueue_notify")
2611 	pscc_t	*pscc;
2612 	psce_t	*psce;
2613 	void		*cookie = NULL;
2614 	int	overrun;
2615 
2616 	ASSERT(MUTEX_HELD(&pm_rsvp_lock));
2617 	switch (cmd) {
2618 	case PSC_PENDING_CHANGE:	/* only for controlling process */
2619 		PMD(PMD_DPM, ("%s: PENDING %s@%s(%s#%d), comp %d, %d -> %d\n",
2620 		    pmf, PM_DEVICE(dip), comp, oldlevel, newlevel))
2621 		psce = pm_psc_dip_to_direct(dip, &pscc);
2622 		ASSERT(psce);
2623 		PMD(PMD_IOCTL, ("%s: PENDING: %s@%s(%s#%d) pm_poll_cnt[%d] "
2624 		    "%d\n", pmf, PM_DEVICE(dip), pscc->pscc_clone,
2625 		    pm_poll_cnt[pscc->pscc_clone]))
2626 		overrun = psc_entry(cmd, psce, dip, comp, newlevel, oldlevel,
2627 		    PSC_DIRECT, canblock);
2628 		PMD(PMD_DPM, ("%s: sig %d\n", pmf, pscc->pscc_clone))
2629 		mutex_enter(&pm_clone_lock);
2630 		if (!overrun)
2631 			pm_poll_cnt[pscc->pscc_clone]++;
2632 		cv_signal(&pm_clones_cv[pscc->pscc_clone]);
2633 		pollwakeup(&pm_pollhead, (POLLRDNORM | POLLIN));
2634 		mutex_exit(&pm_clone_lock);
2635 		break;
2636 	case PSC_HAS_CHANGED:
2637 		PMD(PMD_DPM, ("%s: HAS %s@%s(%s#%d), comp %d, %d -> %d\n",
2638 		    pmf, PM_DEVICE(dip), comp, oldlevel, newlevel))
2639 		if (PM_ISDIRECT(dip) && canblock != PM_CANBLOCK_BYPASS) {
2640 			psce = pm_psc_dip_to_direct(dip, &pscc);
2641 			PMD(PMD_IOCTL, ("%s: HAS: %s@%s(%s#%d) pm_poll_cnt[%d] "
2642 			    "%d\n", pmf, PM_DEVICE(dip), pscc->pscc_clone,
2643 			    pm_poll_cnt[pscc->pscc_clone]))
2644 			overrun = psc_entry(cmd, psce, dip, comp, newlevel,
2645 			    oldlevel, PSC_DIRECT, canblock);
2646 			PMD(PMD_DPM, ("%s: sig %d\n", pmf, pscc->pscc_clone))
2647 			mutex_enter(&pm_clone_lock);
2648 			if (!overrun)
2649 				pm_poll_cnt[pscc->pscc_clone]++;
2650 			cv_signal(&pm_clones_cv[pscc->pscc_clone]);
2651 			pollwakeup(&pm_pollhead, (POLLRDNORM | POLLIN));
2652 			mutex_exit(&pm_clone_lock);
2653 		}
2654 		mutex_enter(&pm_clone_lock);
2655 		rw_enter(&pm_pscc_interest_rwlock, RW_READER);
2656 		while ((psce = psc_interest(&cookie, &pscc)) != NULL) {
2657 			(void) psc_entry(cmd, psce, dip, comp, newlevel,
2658 			    oldlevel, PSC_INTEREST, canblock);
2659 			cv_signal(&pm_clones_cv[pscc->pscc_clone]);
2660 		}
2661 		rw_exit(&pm_pscc_interest_rwlock);
2662 		mutex_exit(&pm_clone_lock);
2663 		break;
2664 #ifdef DEBUG
2665 	default:
2666 		ASSERT(0);
2667 #endif
2668 	}
2669 }
2670 
2671 static void
2672 pm_enqueue_notify_others(pm_ppm_devlist_t **listp, pm_canblock_t canblock)
2673 {
2674 	if (listp) {
2675 		pm_ppm_devlist_t *p, *next = NULL;
2676 
2677 		for (p = *listp; p; p = next) {
2678 			next = p->ppd_next;
2679 			pm_enqueue_notify(PSC_HAS_CHANGED, p->ppd_who,
2680 			    p->ppd_cmpt, p->ppd_new_level, p->ppd_old_level,
2681 			    canblock);
2682 			kmem_free(p, sizeof (pm_ppm_devlist_t));
2683 		}
2684 		*listp = NULL;
2685 	}
2686 }
2687 
2688 /*
2689  * Try to get the power locks of the parent node and target (child)
2690  * node.  Return true if successful (with both locks held) or false
2691  * (with no locks held).
2692  */
2693 static int
2694 pm_try_parent_child_locks(dev_info_t *pdip,
2695     dev_info_t *dip, int *pcircp, int *circp)
2696 {
2697 	if (ndi_devi_tryenter(pdip, pcircp))
2698 		if (PM_TRY_LOCK_POWER(dip, circp)) {
2699 			return (1);
2700 		} else {
2701 			ndi_devi_exit(pdip, *pcircp);
2702 		}
2703 	return (0);
2704 }
2705 
2706 /*
2707  * Determine if the power lock owner is blocked by current thread.
2708  * returns :
2709  * 	1 - If the thread owning the effective power lock (the first lock on
2710  *          which a thread blocks when it does PM_LOCK_POWER) is blocked by
2711  *          a mutex held by the current thread.
2712  *
2713  *	0 - otherwise
2714  *
2715  * Note : This function is called by pm_power_has_changed to determine whether
2716  * it is executing in parallel with pm_set_power.
2717  */
2718 static int
2719 pm_blocked_by_us(dev_info_t *dip)
2720 {
2721 	power_req_t power_req;
2722 	kthread_t *owner;
2723 	int result;
2724 	kmutex_t *mp;
2725 	dev_info_t *ppm = (dev_info_t *)DEVI(dip)->devi_pm_ppm;
2726 
2727 	power_req.request_type = PMR_PPM_POWER_LOCK_OWNER;
2728 	power_req.req.ppm_power_lock_owner_req.who = dip;
2729 	if (pm_ctlops(ppm, dip, DDI_CTLOPS_POWER, &power_req, &result) !=
2730 	    DDI_SUCCESS) {
2731 		/*
2732 		 * It is assumed that if the device is claimed by ppm, ppm
2733 		 * will always implement this request type and it'll always
2734 		 * return success. We panic here, if it fails.
2735 		 */
2736 		panic("pm: Can't determine power lock owner of %s@%s(%s#%d)\n",
2737 		    PM_DEVICE(dip));
2738 		/*NOTREACHED*/
2739 	}
2740 
2741 	if ((owner = power_req.req.ppm_power_lock_owner_req.owner) != NULL &&
2742 	    owner->t_state == TS_SLEEP &&
2743 	    owner->t_sobj_ops &&
2744 	    SOBJ_TYPE(owner->t_sobj_ops) == SOBJ_MUTEX &&
2745 	    (mp = (kmutex_t *)owner->t_wchan) &&
2746 	    mutex_owner(mp) == curthread)
2747 		return (1);
2748 
2749 	return (0);
2750 }
2751 
2752 /*
2753  * Notify parent which wants to hear about a child's power changes.
2754  */
2755 static void
2756 pm_notify_parent(dev_info_t *dip,
2757     dev_info_t *pdip, int comp, int old_level, int level)
2758 {
2759 	pm_bp_has_changed_t bphc;
2760 	pm_sp_misc_t pspm;
2761 	char *pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2762 	int result = DDI_SUCCESS;
2763 
2764 	bphc.bphc_dip = dip;
2765 	bphc.bphc_path = ddi_pathname(dip, pathbuf);
2766 	bphc.bphc_comp = comp;
2767 	bphc.bphc_olevel = old_level;
2768 	bphc.bphc_nlevel = level;
2769 	pspm.pspm_canblock = PM_CANBLOCK_BLOCK;
2770 	pspm.pspm_scan = 0;
2771 	bphc.bphc_private = &pspm;
2772 	(void) (*PM_BUS_POWER_FUNC(pdip))(pdip, NULL,
2773 	    BUS_POWER_HAS_CHANGED, (void *)&bphc, (void *)&result);
2774 	kmem_free(pathbuf, MAXPATHLEN);
2775 }
2776 
2777 /*
2778  * Check if we need to resume a BC device, and make the attach call as required.
2779  */
2780 static int
2781 pm_check_and_resume(dev_info_t *dip, int comp, int old_level, int level)
2782 {
2783 	int ret = DDI_SUCCESS;
2784 
2785 	if (PM_ISBC(dip) && comp == 0 && old_level == 0 && level != 0) {
2786 		ASSERT(DEVI(dip)->devi_pm_flags & PMC_SUSPENDED);
2787 		/* ppm is not interested in DDI_PM_RESUME */
2788 		if ((ret = devi_attach(dip, DDI_PM_RESUME)) != DDI_SUCCESS)
2789 			/* XXX Should we mark it resumed, */
2790 			/* even though it failed? */
2791 			cmn_err(CE_WARN, "!pm: Can't resume %s@%s",
2792 			    PM_NAME(dip), PM_ADDR(dip));
2793 		DEVI(dip)->devi_pm_flags &= ~PMC_SUSPENDED;
2794 	}
2795 
2796 	return (ret);
2797 }
2798 
2799 /*
2800  * Tests outside the lock to see if we should bother to enqueue an entry
2801  * for any watching process.  If yes, then caller will take the lock and
2802  * do the full protocol
2803  */
2804 static int
2805 pm_watchers()
2806 {
2807 	if (pm_processes_stopped)
2808 		return (0);
2809 	return (pm_pscc_direct || pm_pscc_interest);
2810 }
2811 
2812 /*
2813  * A driver is reporting that the power of one of its device's components
2814  * has changed.  Update the power state accordingly.
2815  */
2816 int
2817 pm_power_has_changed(dev_info_t *dip, int comp, int level)
2818 {
2819 	PMD_FUNC(pmf, "pm_power_has_changed")
2820 	int ret;
2821 	dev_info_t *pdip = ddi_get_parent(dip);
2822 	struct pm_component *cp;
2823 	int blocked, circ, pcirc, old_level;
2824 	static int pm_phc_impl(dev_info_t *, int, int, int);
2825 
2826 	if (level < 0) {
2827 		PMD(PMD_FAIL, ("%s: %s@%s(%s#%d): bad level=%d\n", pmf,
2828 		    PM_DEVICE(dip), level))
2829 		return (DDI_FAILURE);
2830 	}
2831 
2832 	PMD(PMD_KIDSUP | PMD_DEP, ("%s: %s@%s(%s#%d), comp=%d, level=%d\n", pmf,
2833 	    PM_DEVICE(dip), comp, level))
2834 
2835 	if (!e_pm_valid_info(dip, NULL) || !e_pm_valid_comp(dip, comp, &cp) ||
2836 	    !e_pm_valid_power(dip, comp, level))
2837 		return (DDI_FAILURE);
2838 
2839 	/*
2840 	 * A driver thread calling pm_power_has_changed and another thread
2841 	 * calling pm_set_power can deadlock.  The problem is not resolvable
2842 	 * by changing lock order, so we use pm_blocked_by_us() to detect
2843 	 * this specific deadlock.  If we can't get the lock immediately
2844 	 * and we are deadlocked, just update the component's level, do
2845 	 * notifications, and return.  We intend to update the total power
2846 	 * state later (if the other thread fails to set power to the
2847 	 * desired level).  If we were called because of a power change on a
2848 	 * component that isn't involved in a set_power op, update all state
2849 	 * immediately.
2850 	 */
2851 	cp = PM_CP(dip, comp);
2852 	while (!pm_try_parent_child_locks(pdip, dip, &pcirc, &circ)) {
2853 		if (((blocked = pm_blocked_by_us(dip)) != 0) &&
2854 		    (cp->pmc_flags & PM_POWER_OP)) {
2855 			if (pm_watchers()) {
2856 				mutex_enter(&pm_rsvp_lock);
2857 				pm_enqueue_notify(PSC_HAS_CHANGED, dip, comp,
2858 				    level, cur_power(cp), PM_CANBLOCK_BLOCK);
2859 				mutex_exit(&pm_rsvp_lock);
2860 			}
2861 			if (pdip && PM_WANTS_NOTIFICATION(pdip))
2862 				pm_notify_parent(dip,
2863 				    pdip, comp, cur_power(cp), level);
2864 			(void) pm_check_and_resume(dip,
2865 			    comp, cur_power(cp), level);
2866 
2867 			/*
2868 			 * Stash the old power index, update curpwr, and flag
2869 			 * that the total power state needs to be synched.
2870 			 */
2871 			cp->pmc_flags |= PM_PHC_WHILE_SET_POWER;
2872 			/*
2873 			 * Several pm_power_has_changed calls could arrive
2874 			 * while the set power path remains blocked.  Keep the
2875 			 * oldest old power and the newest new power of any
2876 			 * sequence of phc calls which arrive during deadlock.
2877 			 */
2878 			if (cp->pmc_phc_pwr == PM_LEVEL_UNKNOWN)
2879 				cp->pmc_phc_pwr = cp->pmc_cur_pwr;
2880 			cp->pmc_cur_pwr =
2881 			    pm_level_to_index(dip, cp, level);
2882 			PMD(PMD_PHC, ("%s: deadlock for %s@%s(%s#%d), comp=%d, "
2883 			    "level=%d\n", pmf, PM_DEVICE(dip), comp, level))
2884 			return (DDI_SUCCESS);
2885 		} else
2886 			if (blocked) {	/* blocked, but different cmpt? */
2887 				if (!ndi_devi_tryenter(pdip, &pcirc)) {
2888 					cmn_err(CE_NOTE,
2889 					    "!pm: parent kuc not updated due "
2890 					    "to possible deadlock.\n");
2891 					return (pm_phc_impl(dip,
2892 						    comp, level, 1));
2893 				}
2894 				old_level = cur_power(cp);
2895 				if (pdip && !PM_WANTS_NOTIFICATION(pdip) &&
2896 				    (!PM_ISBC(dip) || comp == 0) &&
2897 				    POWERING_ON(old_level, level))
2898 					pm_hold_power(pdip);
2899 				ret = pm_phc_impl(dip, comp, level, 1);
2900 				if (pdip && !PM_WANTS_NOTIFICATION(pdip)) {
2901 					if ((!PM_ISBC(dip) ||
2902 					    comp == 0) && level == 0 &&
2903 					    old_level != PM_LEVEL_UNKNOWN)
2904 						pm_rele_power(pdip);
2905 				}
2906 				ndi_devi_exit(pdip, pcirc);
2907 				/* child lock not held: deadlock */
2908 				return (ret);
2909 			}
2910 		delay(1);
2911 		PMD(PMD_PHC, ("%s: try lock again\n", pmf))
2912 	}
2913 
2914 	/* non-deadlock case */
2915 	old_level = cur_power(cp);
2916 	if (pdip && !PM_WANTS_NOTIFICATION(pdip) &&
2917 	    (!PM_ISBC(dip) || comp == 0) && POWERING_ON(old_level, level))
2918 		pm_hold_power(pdip);
2919 	ret = pm_phc_impl(dip, comp, level, 1);
2920 	if (pdip && !PM_WANTS_NOTIFICATION(pdip)) {
2921 		if ((!PM_ISBC(dip) || comp == 0) && level == 0 &&
2922 		    old_level != PM_LEVEL_UNKNOWN)
2923 			pm_rele_power(pdip);
2924 	}
2925 	PM_UNLOCK_POWER(dip, circ);
2926 	ndi_devi_exit(pdip, pcirc);
2927 	return (ret);
2928 }
2929 
2930 /*
2931  * Account for power changes to a component of the the console frame buffer.
2932  * If lowering power from full (or "unkown", which is treatd as full)
2933  * we will increment the "components off" count of the fb device.
2934  * Subsequent lowering of the same component doesn't affect the count.  If
2935  * raising a component back to full power, we will decrement the count.
2936  *
2937  * Return: the increment value for pm_cfb_comps_off (-1, 0, or 1)
2938  */
2939 static int
2940 calc_cfb_comps_incr(dev_info_t *dip, int cmpt, int old, int new)
2941 {
2942 	struct pm_component *cp = PM_CP(dip, cmpt);
2943 	int on = (old == PM_LEVEL_UNKNOWN || old == cp->pmc_norm_pwr);
2944 	int want_normal = (new == cp->pmc_norm_pwr);
2945 	int incr = 0;
2946 
2947 	if (on && !want_normal)
2948 		incr = 1;
2949 	else if (!on && want_normal)
2950 		incr = -1;
2951 	return (incr);
2952 }
2953 
2954 /*
2955  * Adjust the count of console frame buffer components < full power.
2956  */
2957 static void
2958 update_comps_off(int incr, dev_info_t *dip)
2959 {
2960 		mutex_enter(&pm_cfb_lock);
2961 		pm_cfb_comps_off += incr;
2962 		ASSERT(pm_cfb_comps_off <= PM_NUMCMPTS(dip));
2963 		mutex_exit(&pm_cfb_lock);
2964 }
2965 
2966 /*
2967  * Update the power state in the framework (via the ppm).  The 'notify'
2968  * argument tells whether to notify watchers.  Power lock is already held.
2969  */
2970 static int
2971 pm_phc_impl(dev_info_t *dip, int comp, int level, int notify)
2972 {
2973 	PMD_FUNC(pmf, "phc_impl")
2974 	power_req_t power_req;
2975 	int i, dodeps = 0;
2976 	dev_info_t *pdip = ddi_get_parent(dip);
2977 	int result;
2978 	int old_level;
2979 	struct pm_component *cp;
2980 	int incr = 0;
2981 	dev_info_t *ppm = (dev_info_t *)DEVI(dip)->devi_pm_ppm;
2982 	int work_type = 0;
2983 	char *pathbuf;
2984 
2985 	/* Must use "official" power level for this test. */
2986 	cp = PM_CP(dip, comp);
2987 	old_level = (cp->pmc_flags & PM_PHC_WHILE_SET_POWER ?
2988 	    cp->pmc_phc_pwr : cp->pmc_cur_pwr);
2989 	if (old_level != PM_LEVEL_UNKNOWN)
2990 		old_level = cp->pmc_comp.pmc_lvals[old_level];
2991 
2992 	if (level == old_level) {
2993 		PMD(PMD_SET, ("%s: %s@%s(%s#%d), comp=%d is already at "
2994 		    "level=%d\n", pmf, PM_DEVICE(dip), comp, level))
2995 		return (DDI_SUCCESS);
2996 	}
2997 
2998 	/*
2999 	 * Tell ppm about this.
3000 	 */
3001 	power_req.request_type = PMR_PPM_POWER_CHANGE_NOTIFY;
3002 	power_req.req.ppm_notify_level_req.who = dip;
3003 	power_req.req.ppm_notify_level_req.cmpt = comp;
3004 	power_req.req.ppm_notify_level_req.new_level = level;
3005 	power_req.req.ppm_notify_level_req.old_level = old_level;
3006 	if (pm_ctlops(ppm, dip, DDI_CTLOPS_POWER, &power_req,
3007 	    &result) == DDI_FAILURE) {
3008 		PMD(PMD_FAIL, ("%s: pm_ctlops %s@%s(%s#%d) to %d failed\n",
3009 		    pmf, PM_DEVICE(dip), level))
3010 		return (DDI_FAILURE);
3011 	}
3012 
3013 	if (PM_IS_CFB(dip)) {
3014 		incr = calc_cfb_comps_incr(dip, comp, old_level, level);
3015 
3016 		if (incr) {
3017 			update_comps_off(incr, dip);
3018 			PMD(PMD_CFB, ("%s: %s@%s(%s#%d) comp=%d %d->%d "
3019 			    "cfb_comps_off->%d\n", pmf, PM_DEVICE(dip),
3020 			    comp, old_level, level, pm_cfb_comps_off))
3021 		}
3022 	}
3023 	e_pm_set_cur_pwr(dip, PM_CP(dip, comp), level);
3024 	result = DDI_SUCCESS;
3025 
3026 	if (notify) {
3027 		if (pdip && PM_WANTS_NOTIFICATION(pdip))
3028 			pm_notify_parent(dip, pdip, comp, old_level, level);
3029 		(void) pm_check_and_resume(dip, comp, old_level, level);
3030 	}
3031 
3032 	/*
3033 	 * Decrement the dependency kidsup count if we turn a device
3034 	 * off.
3035 	 */
3036 	if (POWERING_OFF(old_level, level)) {
3037 		dodeps = 1;
3038 		for (i = 0; i < PM_NUMCMPTS(dip); i++) {
3039 			cp = PM_CP(dip, i);
3040 			if (cur_power(cp)) {
3041 				dodeps = 0;
3042 				break;
3043 			}
3044 		}
3045 		if (dodeps)
3046 			work_type = PM_DEP_WK_POWER_OFF;
3047 	}
3048 
3049 	/*
3050 	 * Increment if we turn it on. Check to see
3051 	 * if other comps are already on, if so,
3052 	 * dont increment.
3053 	 */
3054 	if (POWERING_ON(old_level, level)) {
3055 		dodeps = 1;
3056 		for (i = 0; i < PM_NUMCMPTS(dip); i++) {
3057 			cp = PM_CP(dip, i);
3058 			if (comp == i)
3059 				continue;
3060 			/* -1 also treated as 0 in this case */
3061 			if (cur_power(cp) > 0) {
3062 				dodeps = 0;
3063 				break;
3064 			}
3065 		}
3066 		if (dodeps)
3067 			work_type = PM_DEP_WK_POWER_ON;
3068 	}
3069 
3070 	if (dodeps) {
3071 		pathbuf = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
3072 		(void) ddi_pathname(dip, pathbuf);
3073 		pm_dispatch_to_dep_thread(work_type, pathbuf, NULL,
3074 		    PM_DEP_NOWAIT, NULL, 0);
3075 		kmem_free(pathbuf, MAXPATHLEN);
3076 	}
3077 
3078 	if (notify && (level != old_level) && pm_watchers()) {
3079 		mutex_enter(&pm_rsvp_lock);
3080 		pm_enqueue_notify(PSC_HAS_CHANGED, dip, comp, level, old_level,
3081 		    PM_CANBLOCK_BLOCK);
3082 		mutex_exit(&pm_rsvp_lock);
3083 	}
3084 
3085 	PMD(PMD_RESCAN, ("%s: %s@%s(%s#%d): pm_rescan\n", pmf, PM_DEVICE(dip)))
3086 	pm_rescan(dip);
3087 	return (DDI_SUCCESS);
3088 }
3089 
3090 /*
3091  * This function is called at startup time to notify pm of the existence
3092  * of any platform power managers for this platform.  As a result of
3093  * this registration, each function provided will be called each time
3094  * a device node is attached, until one returns true, and it must claim the
3095  * device node (by returning non-zero) if it wants to be involved in the
3096  * node's power management.  If it does claim the node, then it will
3097  * subsequently be notified of attach and detach events.
3098  *
3099  */
3100 
3101 int
3102 pm_register_ppm(int (*func)(dev_info_t *), dev_info_t *dip)
3103 {
3104 	PMD_FUNC(pmf, "register_ppm")
3105 	struct ppm_callbacks *ppmcp;
3106 	pm_component_t *cp;
3107 	int i, pwr, result, circ;
3108 	power_req_t power_req;
3109 	struct ppm_notify_level_req *p = &power_req.req.ppm_notify_level_req;
3110 	void pm_ppm_claim(dev_info_t *);
3111 
3112 	mutex_enter(&ppm_lock);
3113 	ppmcp = ppm_callbacks;
3114 	for (i = 0; i < MAX_PPM_HANDLERS; i++, ppmcp++) {
3115 		if (ppmcp->ppmc_func == NULL) {
3116 			ppmcp->ppmc_func = func;
3117 			ppmcp->ppmc_dip = dip;
3118 			break;
3119 		}
3120 	}
3121 	mutex_exit(&ppm_lock);
3122 
3123 	if (i >= MAX_PPM_HANDLERS)
3124 		return (DDI_FAILURE);
3125 	while ((dip = ddi_get_parent(dip)) != NULL) {
3126 		if (PM_GET_PM_INFO(dip) == NULL)
3127 			continue;
3128 		pm_ppm_claim(dip);
3129 		if (pm_ppm_claimed(dip)) {
3130 			/*
3131 			 * Tell ppm about this.
3132 			 */
3133 			power_req.request_type = PMR_PPM_POWER_CHANGE_NOTIFY;
3134 			p->old_level = PM_LEVEL_UNKNOWN;
3135 			p->who = dip;
3136 			PM_LOCK_POWER(dip, &circ);
3137 			for (i = 0; i < PM_NUMCMPTS(dip); i++) {
3138 				cp = PM_CP(dip, i);
3139 				pwr = cp->pmc_cur_pwr;
3140 				if (pwr != PM_LEVEL_UNKNOWN) {
3141 					p->cmpt = i;
3142 					p->new_level = cur_power(cp);
3143 					p->old_level = PM_LEVEL_UNKNOWN;
3144 					if (pm_ctlops(PPM(dip), dip,
3145 					    DDI_CTLOPS_POWER, &power_req,
3146 					    &result) == DDI_FAILURE) {
3147 						PMD(PMD_FAIL, ("%s: pc "
3148 						    "%s@%s(%s#%d) to %d "
3149 						    "fails\n", pmf,
3150 						    PM_DEVICE(dip), pwr))
3151 					}
3152 				}
3153 			}
3154 			PM_UNLOCK_POWER(dip, circ);
3155 		}
3156 	}
3157 	return (DDI_SUCCESS);
3158 }
3159 
3160 /*
3161  * Call the ppm's that have registered and adjust the devinfo struct as
3162  * appropriate.  First one to claim it gets it.  The sets of devices claimed
3163  * by each ppm are assumed to be disjoint.
3164  */
3165 void
3166 pm_ppm_claim(dev_info_t *dip)
3167 {
3168 	struct ppm_callbacks *ppmcp;
3169 
3170 	if (PPM(dip)) {
3171 		return;
3172 	}
3173 	mutex_enter(&ppm_lock);
3174 	for (ppmcp = ppm_callbacks; ppmcp->ppmc_func; ppmcp++) {
3175 		if ((*ppmcp->ppmc_func)(dip)) {
3176 			DEVI(dip)->devi_pm_ppm =
3177 			    (struct dev_info *)ppmcp->ppmc_dip;
3178 			mutex_exit(&ppm_lock);
3179 			return;
3180 		}
3181 	}
3182 	mutex_exit(&ppm_lock);
3183 }
3184 
3185 /*
3186  * Node is being detached so stop autopm until we see if it succeeds, in which
3187  * case pm_stop will be called.  For backwards compatible devices we bring the
3188  * device up to full power on the assumption the detach will succeed.
3189  */
3190 void
3191 pm_detaching(dev_info_t *dip)
3192 {
3193 	PMD_FUNC(pmf, "detaching")
3194 	pm_info_t *info = PM_GET_PM_INFO(dip);
3195 	int iscons;
3196 
3197 	PMD(PMD_REMDEV, ("%s: %s@%s(%s#%d), %d comps\n", pmf, PM_DEVICE(dip),
3198 	    PM_NUMCMPTS(dip)))
3199 	if (info == NULL)
3200 		return;
3201 	ASSERT(DEVI_IS_DETACHING(dip));
3202 	PM_LOCK_DIP(dip);
3203 	info->pmi_dev_pm_state |= PM_DETACHING;
3204 	PM_UNLOCK_DIP(dip);
3205 	if (!PM_ISBC(dip))
3206 		pm_scan_stop(dip);
3207 
3208 	/*
3209 	 * console and old-style devices get brought up when detaching.
3210 	 */
3211 	iscons = PM_IS_CFB(dip);
3212 	if (iscons || PM_ISBC(dip)) {
3213 		(void) pm_all_to_normal(dip, PM_CANBLOCK_BYPASS);
3214 		if (iscons) {
3215 			mutex_enter(&pm_cfb_lock);
3216 			while (cfb_inuse) {
3217 				mutex_exit(&pm_cfb_lock);
3218 				PMD(PMD_CFB, ("%s: delay; cfb_inuse\n", pmf))
3219 				delay(1);
3220 				mutex_enter(&pm_cfb_lock);
3221 			}
3222 			ASSERT(cfb_dip_detaching == NULL);
3223 			ASSERT(cfb_dip);
3224 			cfb_dip_detaching = cfb_dip;	/* case detach fails */
3225 			cfb_dip = NULL;
3226 			mutex_exit(&pm_cfb_lock);
3227 		}
3228 	}
3229 }
3230 
3231 /*
3232  * Node failed to detach.  If it used to be autopm'd, make it so again.
3233  */
3234 void
3235 pm_detach_failed(dev_info_t *dip)
3236 {
3237 	PMD_FUNC(pmf, "detach_failed")
3238 	pm_info_t *info = PM_GET_PM_INFO(dip);
3239 	int pm_all_at_normal(dev_info_t *);
3240 
3241 	if (info == NULL)
3242 		return;
3243 	ASSERT(DEVI_IS_DETACHING(dip));
3244 	if (info->pmi_dev_pm_state & PM_DETACHING) {
3245 		info->pmi_dev_pm_state &= ~PM_DETACHING;
3246 		if (info->pmi_dev_pm_state & PM_ALLNORM_DEFERRED) {
3247 			/* Make sure the operation is still needed */
3248 			if (!pm_all_at_normal(dip)) {
3249 				if (pm_all_to_normal(dip,
3250 				    PM_CANBLOCK_FAIL) != DDI_SUCCESS) {
3251 					PMD(PMD_ERROR, ("%s: could not bring "
3252 					    "%s@%s(%s#%d) to normal\n", pmf,
3253 					    PM_DEVICE(dip)))
3254 				}
3255 			}
3256 			info->pmi_dev_pm_state &= ~PM_ALLNORM_DEFERRED;
3257 		}
3258 	}
3259 	if (!PM_ISBC(dip)) {
3260 		mutex_enter(&pm_scan_lock);
3261 		if (autopm_enabled)
3262 			pm_scan_init(dip);
3263 		mutex_exit(&pm_scan_lock);
3264 		pm_rescan(dip);
3265 	}
3266 }
3267 
3268 /* generic Backwards Compatible component */
3269 static char *bc_names[] = {"off", "on"};
3270 
3271 static pm_comp_t bc_comp = {"unknown", 2, NULL, NULL, &bc_names[0]};
3272 
3273 static void
3274 e_pm_default_levels(dev_info_t *dip, pm_component_t *cp, int norm)
3275 {
3276 	pm_comp_t *pmc;
3277 	pmc = &cp->pmc_comp;
3278 	pmc->pmc_numlevels = 2;
3279 	pmc->pmc_lvals[0] = 0;
3280 	pmc->pmc_lvals[1] = norm;
3281 	e_pm_set_cur_pwr(dip, cp, norm);
3282 }
3283 
3284 static void
3285 e_pm_default_components(dev_info_t *dip, int cmpts)
3286 {
3287 	int i;
3288 	pm_component_t *p = DEVI(dip)->devi_pm_components;
3289 
3290 	p = DEVI(dip)->devi_pm_components;
3291 	for (i = 0; i < cmpts; i++, p++) {
3292 		p->pmc_comp = bc_comp;	/* struct assignment */
3293 		p->pmc_comp.pmc_lvals = kmem_zalloc(2 * sizeof (int),
3294 		    KM_SLEEP);
3295 		p->pmc_comp.pmc_thresh = kmem_alloc(2 * sizeof (int),
3296 		    KM_SLEEP);
3297 		p->pmc_comp.pmc_numlevels = 2;
3298 		p->pmc_comp.pmc_thresh[0] = INT_MAX;
3299 		p->pmc_comp.pmc_thresh[1] = INT_MAX;
3300 	}
3301 }
3302 
3303 /*
3304  * Called from functions that require components to exist already to allow
3305  * for their creation by parsing the pm-components property.
3306  * Device will not be power managed as a result of this call
3307  * No locking needed because we're single threaded by the ndi_devi_enter
3308  * done while attaching, and the device isn't visible until after it has
3309  * attached
3310  */
3311 int
3312 pm_premanage(dev_info_t *dip, int style)
3313 {
3314 	PMD_FUNC(pmf, "premanage")
3315 	pm_comp_t	*pcp, *compp;
3316 	int		cmpts, i, norm, error;
3317 	pm_component_t *p = DEVI(dip)->devi_pm_components;
3318 	pm_comp_t *pm_autoconfig(dev_info_t *, int *);
3319 
3320 	ASSERT(!PM_IAM_LOCKING_DIP(dip));
3321 	/*
3322 	 * If this dip has already been processed, don't mess with it
3323 	 */
3324 	if (DEVI(dip)->devi_pm_flags & PMC_COMPONENTS_DONE)
3325 		return (DDI_SUCCESS);
3326 	if (DEVI(dip)->devi_pm_flags & PMC_COMPONENTS_FAILED) {
3327 		return (DDI_FAILURE);
3328 	}
3329 	/*
3330 	 * Look up pm-components property and create components accordingly
3331 	 * If that fails, fall back to backwards compatibility
3332 	 */
3333 	if ((compp = pm_autoconfig(dip, &error)) == NULL) {
3334 		/*
3335 		 * If error is set, the property existed but was not well formed
3336 		 */
3337 		if (error || (style == PM_STYLE_NEW)) {
3338 			DEVI(dip)->devi_pm_flags |= PMC_COMPONENTS_FAILED;
3339 			return (DDI_FAILURE);
3340 		}
3341 		/*
3342 		 * If they don't have the pm-components property, then we
3343 		 * want the old "no pm until PM_SET_DEVICE_THRESHOLDS ioctl"
3344 		 * behavior driver must have called pm_create_components, and
3345 		 * we need to flesh out dummy components
3346 		 */
3347 		if ((cmpts = PM_NUMCMPTS(dip)) == 0) {
3348 			/*
3349 			 * Not really failure, but we don't want the
3350 			 * caller to treat it as success
3351 			 */
3352 			return (DDI_FAILURE);
3353 		}
3354 		DEVI(dip)->devi_pm_flags |= PMC_BC;
3355 		e_pm_default_components(dip, cmpts);
3356 		for (i = 0; i < cmpts; i++) {
3357 			/*
3358 			 * if normal power not set yet, we don't really know
3359 			 * what *ANY* of the power values are.  If normal
3360 			 * power is set, then we assume for this backwards
3361 			 * compatible case that the values are 0, normal power.
3362 			 */
3363 			norm = pm_get_normal_power(dip, i);
3364 			if (norm == (uint_t)-1) {
3365 				PMD(PMD_ERROR, ("%s: %s@%s(%s#%d)[%d]\n", pmf,
3366 				    PM_DEVICE(dip), i))
3367 				return (DDI_FAILURE);
3368 			}
3369 			/*
3370 			 * Components of BC devices start at their normal power,
3371 			 * so count them to be not at their lowest power.
3372 			 */
3373 			PM_INCR_NOTLOWEST(dip);
3374 			e_pm_default_levels(dip, PM_CP(dip, i), norm);
3375 		}
3376 	} else {
3377 		/*
3378 		 * e_pm_create_components was called from pm_autoconfig(), it
3379 		 * creates components with no descriptions (or known levels)
3380 		 */
3381 		cmpts = PM_NUMCMPTS(dip);
3382 		ASSERT(cmpts != 0);
3383 		pcp = compp;
3384 		p = DEVI(dip)->devi_pm_components;
3385 		for (i = 0; i < cmpts; i++, p++) {
3386 			p->pmc_comp = *pcp++;   /* struct assignment */
3387 			ASSERT(PM_CP(dip, i)->pmc_cur_pwr == 0);
3388 			e_pm_set_cur_pwr(dip, PM_CP(dip, i), PM_LEVEL_UNKNOWN);
3389 		}
3390 		pm_set_device_threshold(dip, pm_system_idle_threshold,
3391 		    PMC_DEF_THRESH);
3392 		kmem_free(compp, cmpts * sizeof (pm_comp_t));
3393 	}
3394 	return (DDI_SUCCESS);
3395 }
3396 
3397 /*
3398  * Called from during or after the device's attach to let us know it is ready
3399  * to play autopm.   Look up the pm model and manage the device accordingly.
3400  * Returns system call errno value.
3401  * If DDI_ATTACH and DDI_DETACH were in same namespace, this would be
3402  * a little cleaner
3403  *
3404  * Called with dip lock held, return with dip lock unheld.
3405  */
3406 
3407 int
3408 e_pm_manage(dev_info_t *dip, int style)
3409 {
3410 	PMD_FUNC(pmf, "e_manage")
3411 	pm_info_t	*info;
3412 	dev_info_t	*pdip = ddi_get_parent(dip);
3413 	int	pm_thresh_specd(dev_info_t *);
3414 	int	count;
3415 	char	*pathbuf;
3416 
3417 	if (pm_premanage(dip, style) != DDI_SUCCESS) {
3418 		return (DDI_FAILURE);
3419 	}
3420 	PMD(PMD_KIDSUP, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
3421 	ASSERT(PM_GET_PM_INFO(dip) == NULL);
3422 	info = kmem_zalloc(sizeof (pm_info_t), KM_SLEEP);
3423 
3424 	/*
3425 	 * Now set up parent's kidsupcnt.  BC nodes are assumed to start
3426 	 * out at their normal power, so they are "up", others start out
3427 	 * unknown, which is effectively "up".  Parent which want notification
3428 	 * get kidsupcnt of 0 always.
3429 	 */
3430 	count = (PM_ISBC(dip)) ? 1 : PM_NUMCMPTS(dip);
3431 	if (count && pdip && !PM_WANTS_NOTIFICATION(pdip))
3432 		e_pm_hold_rele_power(pdip, count);
3433 
3434 	pm_set_pm_info(dip, info);
3435 	/*
3436 	 * Apply any recorded thresholds
3437 	 */
3438 	(void) pm_thresh_specd(dip);
3439 
3440 	/*
3441 	 * Do dependency processing.
3442 	 */
3443 	pathbuf = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
3444 	(void) ddi_pathname(dip, pathbuf);
3445 	pm_dispatch_to_dep_thread(PM_DEP_WK_ATTACH, pathbuf, pathbuf,
3446 	    PM_DEP_NOWAIT, NULL, 0);
3447 	kmem_free(pathbuf, MAXPATHLEN);
3448 
3449 	if (!PM_ISBC(dip)) {
3450 		mutex_enter(&pm_scan_lock);
3451 		if (autopm_enabled) {
3452 			pm_scan_init(dip);
3453 			mutex_exit(&pm_scan_lock);
3454 			pm_rescan(dip);
3455 		} else {
3456 			mutex_exit(&pm_scan_lock);
3457 		}
3458 	}
3459 	return (0);
3460 }
3461 
3462 /*
3463  * This is the obsolete exported interface for a driver to find out its
3464  * "normal" (max) power.
3465  * We only get components destroyed while no power management is
3466  * going on (and the device is detached), so we don't need a mutex here
3467  */
3468 int
3469 pm_get_normal_power(dev_info_t *dip, int comp)
3470 {
3471 
3472 	if (comp >= 0 && comp < PM_NUMCMPTS(dip)) {
3473 		return (PM_CP(dip, comp)->pmc_norm_pwr);
3474 	}
3475 	return (DDI_FAILURE);
3476 }
3477 
3478 /*
3479  * Fetches the current power level.  Return DDI_SUCCESS or DDI_FAILURE.
3480  */
3481 int
3482 pm_get_current_power(dev_info_t *dip, int comp, int *levelp)
3483 {
3484 	if (comp >= 0 && comp < PM_NUMCMPTS(dip)) {
3485 		*levelp = PM_CURPOWER(dip, comp);
3486 		return (DDI_SUCCESS);
3487 	}
3488 	return (DDI_FAILURE);
3489 }
3490 
3491 /*
3492  * Returns current threshold of indicated component
3493  */
3494 static int
3495 cur_threshold(dev_info_t *dip, int comp)
3496 {
3497 	pm_component_t *cp = PM_CP(dip, comp);
3498 	int pwr;
3499 
3500 	if (PM_ISBC(dip)) {
3501 		/*
3502 		 * backwards compatible nodes only have one threshold
3503 		 */
3504 		return (cp->pmc_comp.pmc_thresh[1]);
3505 	}
3506 	pwr = cp->pmc_cur_pwr;
3507 	if (pwr == PM_LEVEL_UNKNOWN) {
3508 		int thresh;
3509 		if (DEVI(dip)->devi_pm_flags & PMC_NEXDEF_THRESH)
3510 			thresh = pm_default_nexus_threshold;
3511 		else
3512 			thresh = pm_system_idle_threshold;
3513 		return (thresh);
3514 	}
3515 	ASSERT(cp->pmc_comp.pmc_thresh);
3516 	return (cp->pmc_comp.pmc_thresh[pwr]);
3517 }
3518 
3519 /*
3520  * Compute next lower component power level given power index.
3521  */
3522 static int
3523 pm_next_lower_power(pm_component_t *cp, int pwrndx)
3524 {
3525 	int nxt_pwr;
3526 
3527 	if (pwrndx == PM_LEVEL_UNKNOWN) {
3528 		nxt_pwr = cp->pmc_comp.pmc_lvals[0];
3529 	} else {
3530 		pwrndx--;
3531 		ASSERT(pwrndx >= 0);
3532 		nxt_pwr = cp->pmc_comp.pmc_lvals[pwrndx];
3533 	}
3534 	return (nxt_pwr);
3535 }
3536 
3537 /*
3538  * Bring all components of device to normal power
3539  */
3540 int
3541 pm_all_to_normal(dev_info_t *dip, pm_canblock_t canblock)
3542 {
3543 	PMD_FUNC(pmf, "all_to_normal")
3544 	int		*normal;
3545 	int		i, ncomps, result;
3546 	size_t		size;
3547 	int		changefailed = 0;
3548 
3549 	PMD(PMD_ALLNORM, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
3550 	ASSERT(PM_GET_PM_INFO(dip));
3551 	if (pm_get_norm_pwrs(dip, &normal, &size) != DDI_SUCCESS) {
3552 		PMD(PMD_ALLNORM, ("%s: can't get norm pwrs for "
3553 		    "%s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
3554 		return (DDI_FAILURE);
3555 	}
3556 	ncomps = PM_NUMCMPTS(dip);
3557 	for (i = 0; i < ncomps; i++) {
3558 		if (pm_set_power(dip, i, normal[i],
3559 		    PM_LEVEL_UPONLY, canblock, 0, &result) != DDI_SUCCESS) {
3560 			changefailed++;
3561 			PMD(PMD_ALLNORM | PMD_FAIL, ("%s: failed to set "
3562 			    "%s@%s(%s#%d)[%d] to %d, errno %d\n", pmf,
3563 			    PM_DEVICE(dip), i, normal[i], result))
3564 		}
3565 	}
3566 	kmem_free(normal, size);
3567 	if (changefailed) {
3568 		PMD(PMD_FAIL, ("%s: failed to set %d comps %s@%s(%s#%d) "
3569 		    "to full power\n", pmf, changefailed, PM_DEVICE(dip)))
3570 		return (DDI_FAILURE);
3571 	}
3572 	return (DDI_SUCCESS);
3573 }
3574 
3575 /*
3576  * Returns true if all components of device are at normal power
3577  */
3578 int
3579 pm_all_at_normal(dev_info_t *dip)
3580 {
3581 	PMD_FUNC(pmf, "all_at_normal")
3582 	int		*normal;
3583 	int		i;
3584 	size_t		size;
3585 
3586 	PMD(PMD_ALLNORM, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
3587 	if (pm_get_norm_pwrs(dip, &normal, &size) != DDI_SUCCESS) {
3588 		PMD(PMD_ALLNORM, ("%s: can't get normal power\n", pmf))
3589 		return (DDI_FAILURE);
3590 	}
3591 	for (i = 0; i < PM_NUMCMPTS(dip); i++) {
3592 		int current = PM_CURPOWER(dip, i);
3593 		if (normal[i] > current) {
3594 			PMD(PMD_ALLNORM, ("%s: %s@%s(%s#%d) comp=%d, "
3595 			    "norm=%d, cur=%d\n", pmf, PM_DEVICE(dip), i,
3596 			    normal[i], current))
3597 			break;
3598 		}
3599 	}
3600 	kmem_free(normal, size);
3601 	if (i != PM_NUMCMPTS(dip)) {
3602 		return (0);
3603 	}
3604 	return (1);
3605 }
3606 
3607 static void
3608 bring_wekeeps_up(char *keeper)
3609 {
3610 	PMD_FUNC(pmf, "bring_wekeeps_up")
3611 	int i;
3612 	pm_pdr_t *dp;
3613 	pm_info_t *wku_info;
3614 	char *kept_path;
3615 	dev_info_t *kept;
3616 	static void bring_pmdep_up(dev_info_t *, int);
3617 
3618 	if (panicstr) {
3619 		return;
3620 	}
3621 	/*
3622 	 * We process the request even if the keeper detaches because
3623 	 * detach processing expects this to increment kidsupcnt of kept.
3624 	 */
3625 	PMD(PMD_BRING, ("%s: keeper= %s\n", pmf, keeper))
3626 	for (dp = pm_dep_head; dp; dp = dp->pdr_next) {
3627 		if (strcmp(dp->pdr_keeper, keeper) != 0)
3628 			continue;
3629 		for (i = 0; i < dp->pdr_kept_count; i++) {
3630 			kept_path = dp->pdr_kept_paths[i];
3631 			if (kept_path == NULL)
3632 				continue;
3633 			ASSERT(kept_path[0] != '\0');
3634 			if ((kept = pm_name_to_dip(kept_path, 1)) == NULL)
3635 				continue;
3636 			wku_info = PM_GET_PM_INFO(kept);
3637 			if (wku_info == NULL) {
3638 				if (kept)
3639 					ddi_release_devi(kept);
3640 				continue;
3641 			}
3642 			/*
3643 			 * Don't mess with it if it is being detached, it isn't
3644 			 * safe to call its power entry point
3645 			 */
3646 			if (wku_info->pmi_dev_pm_state & PM_DETACHING) {
3647 				if (kept)
3648 					ddi_release_devi(kept);
3649 				continue;
3650 			}
3651 			bring_pmdep_up(kept, 1);
3652 			ddi_release_devi(kept);
3653 		}
3654 	}
3655 }
3656 
3657 /*
3658  * Bring up the 'kept' device passed as argument
3659  */
3660 static void
3661 bring_pmdep_up(dev_info_t *kept_dip, int hold)
3662 {
3663 	PMD_FUNC(pmf, "bring_pmdep_up")
3664 	int is_all_at_normal = 0;
3665 
3666 	/*
3667 	 * If the kept device has been unmanaged, do nothing.
3668 	 */
3669 	if (!PM_GET_PM_INFO(kept_dip))
3670 		return;
3671 
3672 	/* Just ignore DIRECT PM device till they are released. */
3673 	if (!pm_processes_stopped && PM_ISDIRECT(kept_dip) &&
3674 	    !(is_all_at_normal = pm_all_at_normal(kept_dip))) {
3675 		PMD(PMD_BRING, ("%s: can't bring up PM_DIRECT %s@%s(%s#%d) "
3676 		    "controlling process did something else\n", pmf,
3677 		    PM_DEVICE(kept_dip)))
3678 		DEVI(kept_dip)->devi_pm_flags |= PMC_SKIP_BRINGUP;
3679 		return;
3680 	}
3681 	/* if we got here the keeper had a transition from OFF->ON */
3682 	if (hold)
3683 		pm_hold_power(kept_dip);
3684 
3685 	if (!is_all_at_normal)
3686 		(void) pm_all_to_normal(kept_dip, PM_CANBLOCK_FAIL);
3687 }
3688 
3689 /*
3690  * A bunch of stuff that belongs only to the next routine (or two)
3691  */
3692 
3693 static const char namestr[] = "NAME=";
3694 static const int nameln = sizeof (namestr) - 1;
3695 static const char pmcompstr[] = "pm-components";
3696 
3697 struct pm_comp_pkg {
3698 	pm_comp_t		*comp;
3699 	struct pm_comp_pkg	*next;
3700 };
3701 
3702 #define	isdigit(ch)	((ch) >= '0' && (ch) <= '9')
3703 
3704 #define	isxdigit(ch)	(isdigit(ch) || ((ch) >= 'a' && (ch) <= 'f') || \
3705 			((ch) >= 'A' && (ch) <= 'F'))
3706 
3707 /*
3708  * Rather than duplicate this code ...
3709  * (this code excerpted from the function that follows it)
3710  */
3711 #define	FINISH_COMP { \
3712 	ASSERT(compp); \
3713 	compp->pmc_lnames_sz = size; \
3714 	tp = compp->pmc_lname_buf = kmem_alloc(size, KM_SLEEP); \
3715 	compp->pmc_numlevels = level; \
3716 	compp->pmc_lnames = kmem_alloc(level * sizeof (char *), KM_SLEEP); \
3717 	compp->pmc_lvals = kmem_alloc(level * sizeof (int), KM_SLEEP); \
3718 	compp->pmc_thresh = kmem_alloc(level * sizeof (int), KM_SLEEP); \
3719 	/* copy string out of prop array into buffer */ \
3720 	for (j = 0; j < level; j++) { \
3721 		compp->pmc_thresh[j] = INT_MAX;		/* only [0] sticks */ \
3722 		compp->pmc_lvals[j] = lvals[j]; \
3723 		(void) strcpy(tp, lnames[j]); \
3724 		compp->pmc_lnames[j] = tp; \
3725 		tp += lszs[j]; \
3726 	} \
3727 	ASSERT(tp > compp->pmc_lname_buf && tp <= \
3728 	    compp->pmc_lname_buf + compp->pmc_lnames_sz); \
3729 	}
3730 
3731 /*
3732  * Create (empty) component data structures.
3733  */
3734 static void
3735 e_pm_create_components(dev_info_t *dip, int num_components)
3736 {
3737 	struct pm_component *compp, *ocompp;
3738 	int i, size = 0;
3739 
3740 	ASSERT(!PM_IAM_LOCKING_DIP(dip));
3741 	ASSERT(!DEVI(dip)->devi_pm_components);
3742 	ASSERT(!(DEVI(dip)->devi_pm_flags & PMC_COMPONENTS_DONE));
3743 	size = sizeof (struct pm_component) * num_components;
3744 
3745 	compp = kmem_zalloc(size, KM_SLEEP);
3746 	ocompp = compp;
3747 	DEVI(dip)->devi_pm_comp_size = size;
3748 	DEVI(dip)->devi_pm_num_components = num_components;
3749 	PM_LOCK_BUSY(dip);
3750 	for (i = 0; i < num_components;  i++) {
3751 		compp->pmc_timestamp = gethrestime_sec();
3752 		compp->pmc_norm_pwr = (uint_t)-1;
3753 		compp++;
3754 	}
3755 	PM_UNLOCK_BUSY(dip);
3756 	DEVI(dip)->devi_pm_components = ocompp;
3757 	DEVI(dip)->devi_pm_flags |= PMC_COMPONENTS_DONE;
3758 }
3759 
3760 /*
3761  * Parse hex or decimal value from char string
3762  */
3763 static char *
3764 pm_parsenum(char *cp, int *valp)
3765 {
3766 	int ch, offset;
3767 	char numbuf[256];
3768 	char *np = numbuf;
3769 	int value = 0;
3770 
3771 	ch = *cp++;
3772 	if (isdigit(ch)) {
3773 		if (ch == '0') {
3774 			if ((ch = *cp++) == 'x' || ch == 'X') {
3775 				ch = *cp++;
3776 				while (isxdigit(ch)) {
3777 					*np++ = (char)ch;
3778 					ch = *cp++;
3779 				}
3780 				*np = 0;
3781 				cp--;
3782 				goto hexval;
3783 			} else {
3784 				goto digit;
3785 			}
3786 		} else {
3787 digit:
3788 			while (isdigit(ch)) {
3789 				*np++ = (char)ch;
3790 				ch = *cp++;
3791 			}
3792 			*np = 0;
3793 			cp--;
3794 			goto decval;
3795 		}
3796 	} else
3797 		return (NULL);
3798 
3799 hexval:
3800 	for (np = numbuf; *np; np++) {
3801 		if (*np >= 'a' && *np <= 'f')
3802 			offset = 'a' - 10;
3803 		else if (*np >= 'A' && *np <= 'F')
3804 			offset = 'A' - 10;
3805 		else if (*np >= '0' && *np <= '9')
3806 			offset = '0';
3807 		value *= 16;
3808 		value += *np - offset;
3809 	}
3810 	*valp = value;
3811 	return (cp);
3812 
3813 decval:
3814 	offset = '0';
3815 	for (np = numbuf; *np; np++) {
3816 		value *= 10;
3817 		value += *np - offset;
3818 	}
3819 	*valp = value;
3820 	return (cp);
3821 }
3822 
3823 /*
3824  * Set max (previously documented as "normal") power.
3825  */
3826 static void
3827 e_pm_set_max_power(dev_info_t *dip, int component_number, int level)
3828 {
3829 	PM_CP(dip, component_number)->pmc_norm_pwr = level;
3830 }
3831 
3832 /*
3833  * Internal routine for destroying components
3834  * It is called even when there might not be any, so it must be forgiving.
3835  */
3836 static void
3837 e_pm_destroy_components(dev_info_t *dip)
3838 {
3839 	int i;
3840 	struct pm_component *cp;
3841 
3842 	ASSERT(!PM_IAM_LOCKING_DIP(dip));
3843 	if (PM_NUMCMPTS(dip) == 0)
3844 		return;
3845 	cp = DEVI(dip)->devi_pm_components;
3846 	ASSERT(cp);
3847 	for (i = 0; i < PM_NUMCMPTS(dip); i++, cp++) {
3848 		int nlevels = cp->pmc_comp.pmc_numlevels;
3849 		kmem_free(cp->pmc_comp.pmc_lvals, nlevels * sizeof (int));
3850 		kmem_free(cp->pmc_comp.pmc_thresh, nlevels * sizeof (int));
3851 		/*
3852 		 * For BC nodes, the rest is static in bc_comp, so skip it
3853 		 */
3854 		if (PM_ISBC(dip))
3855 			continue;
3856 		kmem_free(cp->pmc_comp.pmc_name, cp->pmc_comp.pmc_name_sz);
3857 		kmem_free(cp->pmc_comp.pmc_lnames, nlevels * sizeof (char *));
3858 		kmem_free(cp->pmc_comp.pmc_lname_buf,
3859 				cp->pmc_comp.pmc_lnames_sz);
3860 	}
3861 	kmem_free(DEVI(dip)->devi_pm_components, DEVI(dip)->devi_pm_comp_size);
3862 	DEVI(dip)->devi_pm_components = NULL;
3863 	DEVI(dip)->devi_pm_num_components = 0;
3864 	DEVI(dip)->devi_pm_flags &=
3865 	    ~(PMC_COMPONENTS_DONE | PMC_COMPONENTS_FAILED);
3866 }
3867 
3868 /*
3869  * Read the pm-components property (if there is one) and use it to set up
3870  * components.  Returns a pointer to an array of component structures if
3871  * pm-components found and successfully parsed, else returns NULL.
3872  * Sets error return *errp to true to indicate a failure (as opposed to no
3873  * property being present).
3874  */
3875 pm_comp_t *
3876 pm_autoconfig(dev_info_t *dip, int *errp)
3877 {
3878 	PMD_FUNC(pmf, "autoconfig")
3879 	uint_t nelems;
3880 	char **pp;
3881 	pm_comp_t *compp = NULL;
3882 	int i, j, level, components = 0;
3883 	size_t size = 0;
3884 	struct pm_comp_pkg *p, *ptail;
3885 	struct pm_comp_pkg *phead = NULL;
3886 	int *lvals = NULL;
3887 	int *lszs = NULL;
3888 	int *np = NULL;
3889 	int npi = 0;
3890 	char **lnames = NULL;
3891 	char *cp, *tp;
3892 	pm_comp_t *ret = NULL;
3893 
3894 	ASSERT(!PM_IAM_LOCKING_DIP(dip));
3895 	*errp = 0;	/* assume success */
3896 	if (ddi_prop_lookup_string_array(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
3897 	    (char *)pmcompstr, &pp, &nelems) != DDI_PROP_SUCCESS) {
3898 		return (NULL);
3899 	}
3900 
3901 	if (nelems < 3) {	/* need at least one name and two levels */
3902 		goto errout;
3903 	}
3904 
3905 	/*
3906 	 * pm_create_components is no longer allowed
3907 	 */
3908 	if (PM_NUMCMPTS(dip) != 0) {
3909 		PMD(PMD_ERROR, ("%s: %s@%s(%s#%d) has %d comps\n",
3910 		    pmf, PM_DEVICE(dip), PM_NUMCMPTS(dip)))
3911 		goto errout;
3912 	}
3913 
3914 	lvals = kmem_alloc(nelems * sizeof (int), KM_SLEEP);
3915 	lszs = kmem_alloc(nelems * sizeof (int), KM_SLEEP);
3916 	lnames = kmem_alloc(nelems * sizeof (char *), KM_SLEEP);
3917 	np = kmem_alloc(nelems * sizeof (int), KM_SLEEP);
3918 
3919 	level = 0;
3920 	phead = NULL;
3921 	for (i = 0; i < nelems; i++) {
3922 		cp = pp[i];
3923 		if (!isdigit(*cp)) {	/*  must be name */
3924 			if (strncmp(cp, namestr, nameln) != 0) {
3925 				goto errout;
3926 			}
3927 			if (i != 0) {
3928 				if (level == 0) {	/* no level spec'd */
3929 					PMD(PMD_ERROR, ("%s: no level spec'd\n",
3930 					    pmf))
3931 					goto errout;
3932 				}
3933 				np[npi++] = lvals[level - 1];
3934 				/* finish up previous component levels */
3935 				FINISH_COMP;
3936 			}
3937 			cp += nameln;
3938 			if (!*cp) {
3939 				PMD(PMD_ERROR, ("%s: nsa\n", pmf))
3940 				goto errout;
3941 			}
3942 			p = kmem_zalloc(sizeof (*phead), KM_SLEEP);
3943 			if (phead == NULL) {
3944 				phead = ptail = p;
3945 			} else {
3946 				ptail->next = p;
3947 				ptail = p;
3948 			}
3949 			compp = p->comp = kmem_zalloc(sizeof (pm_comp_t),
3950 			    KM_SLEEP);
3951 			compp->pmc_name_sz = strlen(cp) + 1;
3952 			compp->pmc_name = kmem_zalloc(compp->pmc_name_sz,
3953 			    KM_SLEEP);
3954 			(void) strncpy(compp->pmc_name, cp, compp->pmc_name_sz);
3955 			components++;
3956 			level = 0;
3957 		} else {	/* better be power level <num>=<name> */
3958 #ifdef DEBUG
3959 			tp = cp;
3960 #endif
3961 			if (i == 0 ||
3962 			    (cp = pm_parsenum(cp, &lvals[level])) == NULL) {
3963 				PMD(PMD_ERROR, ("%s: parsenum(%s)\n", pmf, tp))
3964 				goto errout;
3965 			}
3966 #ifdef DEBUG
3967 			tp = cp;
3968 #endif
3969 			if (*cp++ != '=' || !*cp) {
3970 				PMD(PMD_ERROR, ("%s: ex =, got %s\n", pmf, tp))
3971 				goto errout;
3972 			}
3973 
3974 			lszs[level] = strlen(cp) + 1;
3975 			size += lszs[level];
3976 			lnames[level] = cp;	/* points into prop string */
3977 			level++;
3978 		}
3979 	}
3980 	np[npi++] = lvals[level - 1];
3981 	if (level == 0) {	/* ended with a name */
3982 		PMD(PMD_ERROR, ("%s: ewn\n", pmf))
3983 		goto errout;
3984 	}
3985 	FINISH_COMP;
3986 
3987 
3988 	/*
3989 	 * Now we have a list of components--we have to return instead an
3990 	 * array of them, but we can just copy the top level and leave
3991 	 * the rest as is
3992 	 */
3993 	(void) e_pm_create_components(dip, components);
3994 	for (i = 0; i < components; i++)
3995 		e_pm_set_max_power(dip, i, np[i]);
3996 
3997 	ret = kmem_zalloc(components * sizeof (pm_comp_t), KM_SLEEP);
3998 	for (i = 0, p = phead; i < components; i++) {
3999 		ASSERT(p);
4000 		/*
4001 		 * Now sanity-check values:  levels must be monotonically
4002 		 * increasing
4003 		 */
4004 		if (p->comp->pmc_numlevels < 2) {
4005 			PMD(PMD_ERROR, ("%s: comp %s of %s@%s(%s#%d) only %d "
4006 			    "levels\n", pmf,
4007 			    p->comp->pmc_name, PM_DEVICE(dip),
4008 			    p->comp->pmc_numlevels))
4009 			goto errout;
4010 		}
4011 		for (j = 0; j < p->comp->pmc_numlevels; j++) {
4012 			if ((p->comp->pmc_lvals[j] < 0) || ((j > 0) &&
4013 			    (p->comp->pmc_lvals[j] <=
4014 			    p->comp->pmc_lvals[j - 1]))) {
4015 				PMD(PMD_ERROR, ("%s: comp %s of %s@%s(%s#%d) "
4016 				    "not mono. incr, %d follows %d\n", pmf,
4017 				    p->comp->pmc_name, PM_DEVICE(dip),
4018 				    p->comp->pmc_lvals[j],
4019 				    p->comp->pmc_lvals[j - 1]))
4020 				goto errout;
4021 			}
4022 		}
4023 		ret[i] = *p->comp;	/* struct assignment */
4024 		for (j = 0; j < i; j++) {
4025 			/*
4026 			 * Test for unique component names
4027 			 */
4028 			if (strcmp(ret[j].pmc_name, ret[i].pmc_name) == 0) {
4029 				PMD(PMD_ERROR, ("%s: %s of %s@%s(%s#%d) not "
4030 				    "unique\n", pmf, ret[j].pmc_name,
4031 				    PM_DEVICE(dip)))
4032 				goto errout;
4033 			}
4034 		}
4035 		ptail = p;
4036 		p = p->next;
4037 		phead = p;	/* errout depends on phead making sense */
4038 		kmem_free(ptail->comp, sizeof (*ptail->comp));
4039 		kmem_free(ptail, sizeof (*ptail));
4040 	}
4041 out:
4042 	ddi_prop_free(pp);
4043 	if (lvals)
4044 		kmem_free(lvals, nelems * sizeof (int));
4045 	if (lszs)
4046 		kmem_free(lszs, nelems * sizeof (int));
4047 	if (lnames)
4048 		kmem_free(lnames, nelems * sizeof (char *));
4049 	if (np)
4050 		kmem_free(np, nelems * sizeof (int));
4051 	return (ret);
4052 
4053 errout:
4054 	e_pm_destroy_components(dip);
4055 	*errp = 1;	/* signal failure */
4056 	cmn_err(CE_CONT, "!pm: %s property ", pmcompstr);
4057 	for (i = 0; i < nelems - 1; i++)
4058 		cmn_err(CE_CONT, "!'%s', ", pp[i]);
4059 	if (nelems != 0)
4060 		cmn_err(CE_CONT, "!'%s'", pp[nelems - 1]);
4061 	cmn_err(CE_CONT, "! for %s@%s(%s#%d) is ill-formed.\n", PM_DEVICE(dip));
4062 	for (p = phead; p; ) {
4063 		pm_comp_t *pp;
4064 		int n;
4065 
4066 		ptail = p;
4067 		/*
4068 		 * Free component data structures
4069 		 */
4070 		pp = p->comp;
4071 		n = pp->pmc_numlevels;
4072 		if (pp->pmc_name_sz) {
4073 			kmem_free(pp->pmc_name, pp->pmc_name_sz);
4074 		}
4075 		if (pp->pmc_lnames_sz) {
4076 			kmem_free(pp->pmc_lname_buf, pp->pmc_lnames_sz);
4077 		}
4078 		if (pp->pmc_lnames) {
4079 			kmem_free(pp->pmc_lnames, n * (sizeof (char *)));
4080 		}
4081 		if (pp->pmc_thresh) {
4082 			kmem_free(pp->pmc_thresh, n * (sizeof (int)));
4083 		}
4084 		if (pp->pmc_lvals) {
4085 			kmem_free(pp->pmc_lvals, n * (sizeof (int)));
4086 		}
4087 		p = ptail->next;
4088 		kmem_free(ptail, sizeof (*ptail));
4089 	}
4090 	if (ret != NULL)
4091 		kmem_free(ret, components * sizeof (pm_comp_t));
4092 	ret = NULL;
4093 	goto out;
4094 }
4095 
4096 /*
4097  * Set threshold values for a devices components by dividing the target
4098  * threshold (base) by the number of transitions and assign each transition
4099  * that threshold.  This will get the entire device down in the target time if
4100  * all components are idle and even if there are dependencies among components.
4101  *
4102  * Devices may well get powered all the way down before the target time, but
4103  * at least the EPA will be happy.
4104  */
4105 void
4106 pm_set_device_threshold(dev_info_t *dip, int base, int flag)
4107 {
4108 	PMD_FUNC(pmf, "set_device_threshold")
4109 	int target_threshold = (base * 95) / 100;
4110 	int level, comp;		/* loop counters */
4111 	int transitions = 0;
4112 	int ncomp = PM_NUMCMPTS(dip);
4113 	int thresh;
4114 	int remainder;
4115 	pm_comp_t *pmc;
4116 	int i, circ;
4117 
4118 	ASSERT(!PM_IAM_LOCKING_DIP(dip));
4119 	PM_LOCK_DIP(dip);
4120 	/*
4121 	 * First we handle the easy one.  If we're setting the default
4122 	 * threshold for a node with children, then we set it to the
4123 	 * default nexus threshold (currently 0) and mark it as default
4124 	 * nexus threshold instead
4125 	 */
4126 	if (PM_IS_NEXUS(dip)) {
4127 		if (flag == PMC_DEF_THRESH) {
4128 			PMD(PMD_THRESH, ("%s: [%s@%s(%s#%d) NEXDEF]\n", pmf,
4129 			    PM_DEVICE(dip)))
4130 			thresh = pm_default_nexus_threshold;
4131 			for (comp = 0; comp < ncomp; comp++) {
4132 				pmc = &PM_CP(dip, comp)->pmc_comp;
4133 				for (level = 1; level < pmc->pmc_numlevels;
4134 				    level++) {
4135 					pmc->pmc_thresh[level] = thresh;
4136 				}
4137 			}
4138 			DEVI(dip)->devi_pm_dev_thresh =
4139 			    pm_default_nexus_threshold;
4140 			/*
4141 			 * If the nexus node is being reconfigured back to
4142 			 * the default threshold, adjust the notlowest count.
4143 			 */
4144 			if (DEVI(dip)->devi_pm_flags &
4145 			    (PMC_DEV_THRESH|PMC_COMP_THRESH)) {
4146 				PM_LOCK_POWER(dip, &circ);
4147 				for (i = 0; i < PM_NUMCMPTS(dip); i++) {
4148 					if (PM_CURPOWER(dip, i) == 0)
4149 						continue;
4150 					mutex_enter(&pm_compcnt_lock);
4151 					ASSERT(pm_comps_notlowest);
4152 					pm_comps_notlowest--;
4153 					PMD(PMD_LEVEL, ("%s: %s@%s(%s#%d) decr "
4154 					    "notlowest to %d\n", pmf,
4155 					    PM_DEVICE(dip), pm_comps_notlowest))
4156 					if (pm_comps_notlowest == 0)
4157 						pm_ppm_notify_all_lowest(dip,
4158 						    PM_ALL_LOWEST);
4159 					mutex_exit(&pm_compcnt_lock);
4160 				}
4161 				PM_UNLOCK_POWER(dip, circ);
4162 			}
4163 			DEVI(dip)->devi_pm_flags &= PMC_THRESH_NONE;
4164 			DEVI(dip)->devi_pm_flags |= PMC_NEXDEF_THRESH;
4165 			PM_UNLOCK_DIP(dip);
4166 			return;
4167 		} else if (DEVI(dip)->devi_pm_flags & PMC_NEXDEF_THRESH) {
4168 			/*
4169 			 * If the nexus node is being configured for a
4170 			 * non-default threshold, include that node in
4171 			 * the notlowest accounting.
4172 			 */
4173 			PM_LOCK_POWER(dip, &circ);
4174 			for (i = 0; i < PM_NUMCMPTS(dip); i++) {
4175 				if (PM_CURPOWER(dip, i) == 0)
4176 					continue;
4177 				mutex_enter(&pm_compcnt_lock);
4178 				if (pm_comps_notlowest == 0)
4179 					pm_ppm_notify_all_lowest(dip,
4180 					    PM_NOT_ALL_LOWEST);
4181 				pm_comps_notlowest++;
4182 				PMD(PMD_LEVEL, ("%s: %s@%s(%s#%d) incr "
4183 				    "notlowest to %d\n", pmf,
4184 				    PM_DEVICE(dip), pm_comps_notlowest))
4185 				mutex_exit(&pm_compcnt_lock);
4186 			}
4187 			PM_UNLOCK_POWER(dip, circ);
4188 		}
4189 	}
4190 	/*
4191 	 * Compute the total number of transitions for all components
4192 	 * of the device.  Distribute the threshold evenly over them
4193 	 */
4194 	for (comp = 0; comp < ncomp; comp++) {
4195 		pmc = &PM_CP(dip, comp)->pmc_comp;
4196 		ASSERT(pmc->pmc_numlevels > 1);
4197 		transitions += pmc->pmc_numlevels - 1;
4198 	}
4199 	ASSERT(transitions);
4200 	thresh = target_threshold / transitions;
4201 
4202 	for (comp = 0; comp < ncomp; comp++) {
4203 		pmc = &PM_CP(dip, comp)->pmc_comp;
4204 		for (level = 1; level < pmc->pmc_numlevels; level++) {
4205 			pmc->pmc_thresh[level] = thresh;
4206 		}
4207 	}
4208 
4209 #ifdef DEBUG
4210 	for (comp = 0; comp < ncomp; comp++) {
4211 		pmc = &PM_CP(dip, comp)->pmc_comp;
4212 		for (level = 1; level < pmc->pmc_numlevels; level++) {
4213 			PMD(PMD_THRESH, ("%s: thresh before %s@%s(%s#%d) "
4214 			    "comp=%d, level=%d, %d\n", pmf, PM_DEVICE(dip),
4215 			    comp, level, pmc->pmc_thresh[level]))
4216 		}
4217 	}
4218 #endif
4219 	/*
4220 	 * Distribute any remainder till they are all gone
4221 	 */
4222 	remainder = target_threshold - thresh * transitions;
4223 	level = 1;
4224 #ifdef DEBUG
4225 	PMD(PMD_THRESH, ("%s: remainder=%d target_threshold=%d thresh=%d "
4226 	    "trans=%d\n", pmf, remainder, target_threshold, thresh,
4227 	    transitions))
4228 #endif
4229 	while (remainder > 0) {
4230 		comp = 0;
4231 		while (remainder && (comp < ncomp)) {
4232 			pmc = &PM_CP(dip, comp)->pmc_comp;
4233 			if (level < pmc->pmc_numlevels) {
4234 				pmc->pmc_thresh[level] += 1;
4235 				remainder--;
4236 			}
4237 			comp++;
4238 		}
4239 		level++;
4240 	}
4241 #ifdef DEBUG
4242 	for (comp = 0; comp < ncomp; comp++) {
4243 		pmc = &PM_CP(dip, comp)->pmc_comp;
4244 		for (level = 1; level < pmc->pmc_numlevels; level++) {
4245 			PMD(PMD_THRESH, ("%s: thresh after %s@%s(%s#%d) "
4246 			    "comp=%d level=%d, %d\n", pmf, PM_DEVICE(dip),
4247 			    comp, level, pmc->pmc_thresh[level]))
4248 		}
4249 	}
4250 #endif
4251 	ASSERT(PM_IAM_LOCKING_DIP(dip));
4252 	DEVI(dip)->devi_pm_dev_thresh = base;
4253 	DEVI(dip)->devi_pm_flags &= PMC_THRESH_NONE;
4254 	DEVI(dip)->devi_pm_flags |= flag;
4255 	PM_UNLOCK_DIP(dip);
4256 }
4257 
4258 /*
4259  * Called when there is no old-style platform power management driver
4260  */
4261 static int
4262 ddi_no_platform_power(power_req_t *req)
4263 {
4264 	_NOTE(ARGUNUSED(req))
4265 	return (DDI_FAILURE);
4266 }
4267 
4268 /*
4269  * This function calls the entry point supplied by the platform-specific
4270  * pm driver to bring the device component 'pm_cmpt' to power level 'pm_level'.
4271  * The use of global for getting the  function name from platform-specific
4272  * pm driver is not ideal, but it is simple and efficient.
4273  * The previous property lookup was being done in the idle loop on swift
4274  * systems without pmc chips and hurt deskbench performance as well as
4275  * violating scheduler locking rules
4276  */
4277 int	(*pm_platform_power)(power_req_t *) = ddi_no_platform_power;
4278 
4279 /*
4280  * Old obsolete interface for a device to request a power change (but only
4281  * an increase in power)
4282  */
4283 int
4284 ddi_dev_is_needed(dev_info_t *dip, int cmpt, int level)
4285 {
4286 	return (pm_raise_power(dip, cmpt, level));
4287 }
4288 
4289 /*
4290  * The old obsolete interface to platform power management.  Only used by
4291  * Gypsy platform and APM on X86.
4292  */
4293 int
4294 ddi_power(dev_info_t *dip, int pm_cmpt, int pm_level)
4295 {
4296 	power_req_t	request;
4297 
4298 	request.request_type = PMR_SET_POWER;
4299 	request.req.set_power_req.who = dip;
4300 	request.req.set_power_req.cmpt = pm_cmpt;
4301 	request.req.set_power_req.level = pm_level;
4302 	return (ddi_ctlops(dip, dip, DDI_CTLOPS_POWER, &request, NULL));
4303 }
4304 
4305 /*
4306  * A driver can invoke this from its detach routine when DDI_SUSPEND is
4307  * passed.  Returns true if subsequent processing could result in power being
4308  * removed from the device.  The arg is not currently used because it is
4309  * implicit in the operation of cpr/DR.
4310  */
4311 int
4312 ddi_removing_power(dev_info_t *dip)
4313 {
4314 	_NOTE(ARGUNUSED(dip))
4315 	return (pm_powering_down);
4316 }
4317 
4318 /*
4319  * Returns true if a device indicates that its parent handles suspend/resume
4320  * processing for it.
4321  */
4322 int
4323 e_ddi_parental_suspend_resume(dev_info_t *dip)
4324 {
4325 	return (DEVI(dip)->devi_pm_flags & PMC_PARENTAL_SR);
4326 }
4327 
4328 /*
4329  * Called for devices which indicate that their parent does suspend/resume
4330  * handling for them
4331  */
4332 int
4333 e_ddi_suspend(dev_info_t *dip, ddi_detach_cmd_t cmd)
4334 {
4335 	power_req_t	request;
4336 	request.request_type = PMR_SUSPEND;
4337 	request.req.suspend_req.who = dip;
4338 	request.req.suspend_req.cmd = cmd;
4339 	return (ddi_ctlops(dip, dip, DDI_CTLOPS_POWER, &request, NULL));
4340 }
4341 
4342 /*
4343  * Called for devices which indicate that their parent does suspend/resume
4344  * handling for them
4345  */
4346 int
4347 e_ddi_resume(dev_info_t *dip, ddi_attach_cmd_t cmd)
4348 {
4349 	power_req_t	request;
4350 	request.request_type = PMR_RESUME;
4351 	request.req.resume_req.who = dip;
4352 	request.req.resume_req.cmd = cmd;
4353 	return (ddi_ctlops(dip, dip, DDI_CTLOPS_POWER, &request, NULL));
4354 }
4355 
4356 /*
4357  * Old obsolete exported interface for drivers to create components.
4358  * This is now handled by exporting the pm-components property.
4359  */
4360 int
4361 pm_create_components(dev_info_t *dip, int num_components)
4362 {
4363 	PMD_FUNC(pmf, "pm_create_components")
4364 
4365 	if (num_components < 1)
4366 		return (DDI_FAILURE);
4367 
4368 	if (!DEVI_IS_ATTACHING(dip)) {
4369 		return (DDI_FAILURE);
4370 	}
4371 
4372 	/* don't need to lock dip because attach is single threaded */
4373 	if (DEVI(dip)->devi_pm_components) {
4374 		PMD(PMD_ERROR, ("%s: %s@%s(%s#%d) already has %d\n", pmf,
4375 		    PM_DEVICE(dip), PM_NUMCMPTS(dip)))
4376 		return (DDI_FAILURE);
4377 	}
4378 	e_pm_create_components(dip, num_components);
4379 	DEVI(dip)->devi_pm_flags |= PMC_BC;
4380 	e_pm_default_components(dip, num_components);
4381 	return (DDI_SUCCESS);
4382 }
4383 
4384 /*
4385  * Obsolete interface previously called by drivers to destroy their components
4386  * at detach time.  This is now done automatically.  However, we need to keep
4387  * this for the old drivers.
4388  */
4389 void
4390 pm_destroy_components(dev_info_t *dip)
4391 {
4392 	PMD_FUNC(pmf, "pm_destroy_components")
4393 	dev_info_t *pdip = ddi_get_parent(dip);
4394 
4395 	PMD(PMD_REMDEV | PMD_KIDSUP, ("%s: %s@%s(%s#%d)\n", pmf,
4396 	    PM_DEVICE(dip)))
4397 	ASSERT(DEVI_IS_DETACHING(dip));
4398 #ifdef DEBUG
4399 	if (!PM_ISBC(dip))
4400 		cmn_err(CE_WARN, "!driver exporting pm-components property "
4401 		    "(%s@%s) calls pm_destroy_components", PM_NAME(dip),
4402 		    PM_ADDR(dip));
4403 #endif
4404 	/*
4405 	 * We ignore this unless this is an old-style driver, except for
4406 	 * printing the message above
4407 	 */
4408 	if (PM_NUMCMPTS(dip) == 0 || !PM_ISBC(dip)) {
4409 		PMD(PMD_REMDEV, ("%s: ignore %s@%s(%s#%d)\n", pmf,
4410 		    PM_DEVICE(dip)))
4411 		return;
4412 	}
4413 	ASSERT(PM_GET_PM_INFO(dip));
4414 
4415 	/*
4416 	 * pm_unmanage will clear info pointer later, after dealing with
4417 	 * dependencies
4418 	 */
4419 	ASSERT(!PM_GET_PM_SCAN(dip));	/* better be gone already */
4420 	/*
4421 	 * Now adjust parent's kidsupcnt.  We check only comp 0.
4422 	 * Parents that get notification are not adjusted because their
4423 	 * kidsupcnt is always 0 (or 1 during probe and attach).
4424 	 */
4425 	if ((PM_CURPOWER(dip, 0) != 0) && pdip && !PM_WANTS_NOTIFICATION(pdip))
4426 		pm_rele_power(pdip);
4427 #ifdef DEBUG
4428 	else {
4429 		PMD(PMD_KIDSUP, ("%s: kuc stays %s@%s(%s#%d) comps gone\n",
4430 		    pmf, PM_DEVICE(dip)))
4431 	}
4432 #endif
4433 	e_pm_destroy_components(dip);
4434 	/*
4435 	 * Forget we ever knew anything about the components of this  device
4436 	 */
4437 	DEVI(dip)->devi_pm_flags &=
4438 	    ~(PMC_BC | PMC_COMPONENTS_DONE | PMC_COMPONENTS_FAILED);
4439 }
4440 
4441 /*
4442  * Exported interface for a driver to set a component busy.
4443  */
4444 int
4445 pm_busy_component(dev_info_t *dip, int cmpt)
4446 {
4447 	struct pm_component *cp;
4448 
4449 	ASSERT(dip != NULL);
4450 	if (!e_pm_valid_info(dip, NULL) || !e_pm_valid_comp(dip, cmpt, &cp))
4451 		return (DDI_FAILURE);
4452 	PM_LOCK_BUSY(dip);
4453 	cp->pmc_busycount++;
4454 	cp->pmc_timestamp = 0;
4455 	PM_UNLOCK_BUSY(dip);
4456 	return (DDI_SUCCESS);
4457 }
4458 
4459 /*
4460  * Exported interface for a driver to set a component idle.
4461  */
4462 int
4463 pm_idle_component(dev_info_t *dip, int cmpt)
4464 {
4465 	PMD_FUNC(pmf, "pm_idle_component")
4466 	struct pm_component *cp;
4467 	pm_scan_t	*scanp = PM_GET_PM_SCAN(dip);
4468 
4469 	if (!e_pm_valid_info(dip, NULL) || !e_pm_valid_comp(dip, cmpt, &cp))
4470 		return (DDI_FAILURE);
4471 
4472 	PM_LOCK_BUSY(dip);
4473 	if (cp->pmc_busycount) {
4474 		if (--(cp->pmc_busycount) == 0)
4475 			cp->pmc_timestamp = gethrestime_sec();
4476 	} else {
4477 		cp->pmc_timestamp = gethrestime_sec();
4478 	}
4479 
4480 	PM_UNLOCK_BUSY(dip);
4481 
4482 	/*
4483 	 * if device becomes idle during idle down period, try scan it down
4484 	 */
4485 	if (scanp && PM_IS_PID(dip)) {
4486 		PMD(PMD_IDLEDOWN, ("%s: %s@%s(%s#%d) idle.\n", pmf,
4487 		    PM_DEVICE(dip)))
4488 		pm_rescan(dip);
4489 		return (DDI_SUCCESS);
4490 	}
4491 
4492 	/*
4493 	 * handle scan not running with nexus threshold == 0
4494 	 */
4495 
4496 	if (PM_IS_NEXUS(dip) && (cp->pmc_busycount == 0)) {
4497 		pm_rescan(dip);
4498 	}
4499 
4500 	return (DDI_SUCCESS);
4501 }
4502 
4503 /*
4504  * This is the old  obsolete interface called by drivers to set their normal
4505  * power.  Thus we can't fix its behavior or return a value.
4506  * This functionality is replaced by the pm-component property.
4507  * We'll only get components destroyed while no power management is
4508  * going on (and the device is detached), so we don't need a mutex here
4509  */
4510 void
4511 pm_set_normal_power(dev_info_t *dip, int comp, int level)
4512 {
4513 	PMD_FUNC(pmf, "set_normal_power")
4514 #ifdef DEBUG
4515 	if (!PM_ISBC(dip))
4516 		cmn_err(CE_WARN, "!call to pm_set_normal_power() by %s@%s "
4517 		    "(driver exporting pm-components property) ignored",
4518 		    PM_NAME(dip), PM_ADDR(dip));
4519 #endif
4520 	if (PM_ISBC(dip)) {
4521 		PMD(PMD_NORM, ("%s: %s@%s(%s#%d) set normal power comp=%d, "
4522 		    "level=%d\n", pmf, PM_DEVICE(dip), comp, level))
4523 		e_pm_set_max_power(dip, comp, level);
4524 		e_pm_default_levels(dip, PM_CP(dip, comp), level);
4525 	}
4526 }
4527 
4528 /*
4529  * Called on a successfully detached driver to free pm resources
4530  */
4531 static void
4532 pm_stop(dev_info_t *dip)
4533 {
4534 	PMD_FUNC(pmf, "stop")
4535 	dev_info_t *pdip = ddi_get_parent(dip);
4536 
4537 	ASSERT(!PM_IAM_LOCKING_DIP(dip));
4538 	/* stopping scan, destroy scan data structure */
4539 	if (!PM_ISBC(dip)) {
4540 		pm_scan_stop(dip);
4541 		pm_scan_fini(dip);
4542 	}
4543 
4544 	if (PM_GET_PM_INFO(dip) != NULL) {
4545 		if (pm_unmanage(dip) == DDI_SUCCESS) {
4546 			/*
4547 			 * Old style driver may have called
4548 			 * pm_destroy_components already, but just in case ...
4549 			 */
4550 			e_pm_destroy_components(dip);
4551 		} else {
4552 			PMD(PMD_FAIL, ("%s: can't pm_unmanage %s@%s(%s#%d)\n",
4553 			    pmf, PM_DEVICE(dip)))
4554 		}
4555 	} else {
4556 		if (PM_NUMCMPTS(dip))
4557 			e_pm_destroy_components(dip);
4558 		else {
4559 			if (DEVI(dip)->devi_pm_flags & PMC_NOPMKID) {
4560 				DEVI(dip)->devi_pm_flags &= ~PMC_NOPMKID;
4561 				if (pdip && !PM_WANTS_NOTIFICATION(pdip)) {
4562 					pm_rele_power(pdip);
4563 				} else if (pdip &&
4564 				    MDI_VHCI(pdip) && MDI_CLIENT(dip)) {
4565 					(void) mdi_power(pdip,
4566 					    MDI_PM_RELE_POWER,
4567 					    (void *)dip, NULL, 0);
4568 				}
4569 			}
4570 		}
4571 	}
4572 }
4573 
4574 /*
4575  * The node is the subject of a reparse pm props ioctl. Throw away the old
4576  * info and start over.
4577  */
4578 int
4579 e_new_pm_props(dev_info_t *dip)
4580 {
4581 	if (PM_GET_PM_INFO(dip) != NULL) {
4582 		pm_stop(dip);
4583 
4584 		if (e_pm_manage(dip, PM_STYLE_NEW) != DDI_SUCCESS) {
4585 			return (DDI_FAILURE);
4586 		}
4587 	}
4588 	e_pm_props(dip);
4589 	return (DDI_SUCCESS);
4590 }
4591 
4592 /*
4593  * Device has been attached, so process its pm properties
4594  */
4595 void
4596 e_pm_props(dev_info_t *dip)
4597 {
4598 	char *pp;
4599 	int len;
4600 	int flags = 0;
4601 	int propflag = DDI_PROP_DONTPASS|DDI_PROP_CANSLEEP;
4602 
4603 	/*
4604 	 * It doesn't matter if we do this more than once, we should always
4605 	 * get the same answers, and if not, then the last one in is the
4606 	 * best one.
4607 	 */
4608 	if (ddi_getlongprop(DDI_DEV_T_ANY, dip, propflag, "pm-hardware-state",
4609 	    (caddr_t)&pp, &len) == DDI_PROP_SUCCESS) {
4610 		if (strcmp(pp, "needs-suspend-resume") == 0) {
4611 			flags = PMC_NEEDS_SR;
4612 		} else if (strcmp(pp, "no-suspend-resume") == 0) {
4613 			flags = PMC_NO_SR;
4614 		} else if (strcmp(pp, "parental-suspend-resume") == 0) {
4615 			flags = PMC_PARENTAL_SR;
4616 		} else {
4617 			cmn_err(CE_NOTE, "!device %s@%s has unrecognized "
4618 			    "%s property value '%s'", PM_NAME(dip),
4619 			    PM_ADDR(dip), "pm-hardware-state", pp);
4620 		}
4621 		kmem_free(pp, len);
4622 	}
4623 	/*
4624 	 * This next segment (PMC_WANTS_NOTIFY) is in
4625 	 * support of nexus drivers which will want to be involved in
4626 	 * (or at least notified of) their child node's power level transitions.
4627 	 * "pm-want-child-notification?" is defined by the parent.
4628 	 */
4629 	if (ddi_prop_exists(DDI_DEV_T_ANY, dip, propflag,
4630 	    "pm-want-child-notification?") && PM_HAS_BUS_POWER(dip))
4631 		flags |= PMC_WANTS_NOTIFY;
4632 	ASSERT(PM_HAS_BUS_POWER(dip) || !ddi_prop_exists(DDI_DEV_T_ANY,
4633 	    dip, propflag, "pm-want-child-notification?"));
4634 	if (ddi_prop_exists(DDI_DEV_T_ANY, dip, propflag,
4635 	    "no-involuntary-power-cycles"))
4636 		flags |= PMC_NO_INVOL;
4637 	/* devfs single threads us */
4638 	DEVI(dip)->devi_pm_flags |= flags;
4639 }
4640 
4641 /*
4642  * This is the DDI_CTLOPS_POWER handler that is used when there is no ppm
4643  * driver which has claimed a node.
4644  * Sets old_power in arg struct.
4645  */
4646 static int
4647 pm_default_ctlops(dev_info_t *dip, dev_info_t *rdip,
4648     ddi_ctl_enum_t ctlop, void *arg, void *result)
4649 {
4650 	_NOTE(ARGUNUSED(dip))
4651 	PMD_FUNC(pmf, "ctlops")
4652 	power_req_t *reqp = (power_req_t *)arg;
4653 	int retval;
4654 	dev_info_t *target_dip;
4655 	int new_level, old_level, cmpt;
4656 #ifdef DEBUG
4657 	char *format;
4658 #endif
4659 
4660 	/*
4661 	 * The interface for doing the actual power level changes is now
4662 	 * through the DDI_CTLOPS_POWER bus_ctl, so that we can plug in
4663 	 * different platform-specific power control drivers.
4664 	 *
4665 	 * This driver implements the "default" version of this interface.
4666 	 * If no ppm driver has been installed then this interface is called
4667 	 * instead.
4668 	 */
4669 	ASSERT(dip == NULL);
4670 	switch (ctlop) {
4671 	case DDI_CTLOPS_POWER:
4672 		switch (reqp->request_type) {
4673 		case PMR_PPM_SET_POWER:
4674 		{
4675 			target_dip = reqp->req.ppm_set_power_req.who;
4676 			ASSERT(target_dip == rdip);
4677 			new_level = reqp->req.ppm_set_power_req.new_level;
4678 			cmpt = reqp->req.ppm_set_power_req.cmpt;
4679 			/* pass back old power for the PM_LEVEL_UNKNOWN case */
4680 			old_level = PM_CURPOWER(target_dip, cmpt);
4681 			reqp->req.ppm_set_power_req.old_level = old_level;
4682 			retval = pm_power(target_dip, cmpt, new_level);
4683 			PMD(PMD_PPM, ("%s: PPM_SET_POWER %s@%s(%s#%d)[%d] %d->"
4684 			    "%d %s\n", pmf, PM_DEVICE(target_dip), cmpt,
4685 			    old_level, new_level, (retval == DDI_SUCCESS ?
4686 			    "chd" : "no chg")))
4687 			return (retval);
4688 		}
4689 
4690 		case PMR_PPM_PRE_DETACH:
4691 		case PMR_PPM_POST_DETACH:
4692 		case PMR_PPM_PRE_ATTACH:
4693 		case PMR_PPM_POST_ATTACH:
4694 		case PMR_PPM_PRE_PROBE:
4695 		case PMR_PPM_POST_PROBE:
4696 		case PMR_PPM_PRE_RESUME:
4697 		case PMR_PPM_INIT_CHILD:
4698 		case PMR_PPM_UNINIT_CHILD:
4699 #ifdef DEBUG
4700 			switch (reqp->request_type) {
4701 				case PMR_PPM_PRE_DETACH:
4702 					format = "%s: PMR_PPM_PRE_DETACH "
4703 					    "%s@%s(%s#%d)\n";
4704 					break;
4705 				case PMR_PPM_POST_DETACH:
4706 					format = "%s: PMR_PPM_POST_DETACH "
4707 					    "%s@%s(%s#%d) rets %d\n";
4708 					break;
4709 				case PMR_PPM_PRE_ATTACH:
4710 					format = "%s: PMR_PPM_PRE_ATTACH "
4711 					    "%s@%s(%s#%d)\n";
4712 					break;
4713 				case PMR_PPM_POST_ATTACH:
4714 					format = "%s: PMR_PPM_POST_ATTACH "
4715 					    "%s@%s(%s#%d) rets %d\n";
4716 					break;
4717 				case PMR_PPM_PRE_PROBE:
4718 					format = "%s: PMR_PPM_PRE_PROBE "
4719 					    "%s@%s(%s#%d)\n";
4720 					break;
4721 				case PMR_PPM_POST_PROBE:
4722 					format = "%s: PMR_PPM_POST_PROBE "
4723 					    "%s@%s(%s#%d) rets %d\n";
4724 					break;
4725 				case PMR_PPM_PRE_RESUME:
4726 					format = "%s: PMR_PPM_PRE_RESUME "
4727 					    "%s@%s(%s#%d) rets %d\n";
4728 					break;
4729 				case PMR_PPM_INIT_CHILD:
4730 					format = "%s: PMR_PPM_INIT_CHILD "
4731 					    "%s@%s(%s#%d)\n";
4732 					break;
4733 				case PMR_PPM_UNINIT_CHILD:
4734 					format = "%s: PMR_PPM_UNINIT_CHILD "
4735 					    "%s@%s(%s#%d)\n";
4736 					break;
4737 				default:
4738 					break;
4739 			}
4740 			PMD(PMD_PPM, (format, pmf, PM_DEVICE(rdip),
4741 			    reqp->req.ppm_config_req.result))
4742 #endif
4743 			return (DDI_SUCCESS);
4744 
4745 		case PMR_PPM_POWER_CHANGE_NOTIFY:
4746 			/*
4747 			 * Nothing for us to do
4748 			 */
4749 			ASSERT(reqp->req.ppm_notify_level_req.who == rdip);
4750 			PMD(PMD_PPM, ("%s: PMR_PPM_POWER_CHANGE_NOTIFY "
4751 			    "%s@%s(%s#%d)[%d] %d->%d\n", pmf,
4752 			    PM_DEVICE(reqp->req.ppm_notify_level_req.who),
4753 			    reqp->req.ppm_notify_level_req.cmpt,
4754 			    PM_CURPOWER(reqp->req.ppm_notify_level_req.who,
4755 			    reqp->req.ppm_notify_level_req.cmpt),
4756 			    reqp->req.ppm_notify_level_req.new_level))
4757 			return (DDI_SUCCESS);
4758 
4759 		case PMR_PPM_UNMANAGE:
4760 			PMD(PMD_PPM, ("%s: PMR_PPM_UNMANAGE %s@%s(%s#%d)\n",
4761 			    pmf, PM_DEVICE(rdip)))
4762 			return (DDI_SUCCESS);
4763 
4764 		case PMR_PPM_LOCK_POWER:
4765 			pm_lock_power_single(reqp->req.ppm_lock_power_req.who,
4766 			    reqp->req.ppm_lock_power_req.circp);
4767 			return (DDI_SUCCESS);
4768 
4769 		case PMR_PPM_UNLOCK_POWER:
4770 			pm_unlock_power_single(
4771 			    reqp->req.ppm_unlock_power_req.who,
4772 			    reqp->req.ppm_unlock_power_req.circ);
4773 			return (DDI_SUCCESS);
4774 
4775 		case PMR_PPM_TRY_LOCK_POWER:
4776 			*(int *)result = pm_try_locking_power_single(
4777 			    reqp->req.ppm_lock_power_req.who,
4778 			    reqp->req.ppm_lock_power_req.circp);
4779 			return (DDI_SUCCESS);
4780 
4781 		case PMR_PPM_POWER_LOCK_OWNER:
4782 			target_dip = reqp->req.ppm_power_lock_owner_req.who;
4783 			ASSERT(target_dip == rdip);
4784 			reqp->req.ppm_power_lock_owner_req.owner =
4785 			    DEVI(rdip)->devi_busy_thread;
4786 			return (DDI_SUCCESS);
4787 		default:
4788 			PMD(PMD_ERROR, ("%s: default!\n", pmf))
4789 			return (DDI_FAILURE);
4790 		}
4791 
4792 	default:
4793 		PMD(PMD_ERROR, ("%s: unknown\n", pmf))
4794 		return (DDI_FAILURE);
4795 	}
4796 }
4797 
4798 /*
4799  * We overload the bus_ctl ops here--perhaps we ought to have a distinct
4800  * power_ops struct for this functionality instead?
4801  * However, we only ever do this on a ppm driver.
4802  */
4803 int
4804 pm_ctlops(dev_info_t *d, dev_info_t *r, ddi_ctl_enum_t op, void *a, void *v)
4805 {
4806 	int (*fp)();
4807 
4808 	/* if no ppm handler, call the default routine */
4809 	if (d == NULL) {
4810 		return (pm_default_ctlops(d, r, op, a, v));
4811 	}
4812 	if (!d || !r)
4813 		return (DDI_FAILURE);
4814 	ASSERT(DEVI(d)->devi_ops && DEVI(d)->devi_ops->devo_bus_ops &&
4815 		DEVI(d)->devi_ops->devo_bus_ops->bus_ctl);
4816 
4817 	fp = DEVI(d)->devi_ops->devo_bus_ops->bus_ctl;
4818 	return ((*fp)(d, r, op, a, v));
4819 }
4820 
4821 /*
4822  * Called on a node when attach completes or the driver makes its first pm
4823  * call (whichever comes first).
4824  * In the attach case, device may not be power manageable at all.
4825  * Don't need to lock the dip because we're single threaded by the devfs code
4826  */
4827 static int
4828 pm_start(dev_info_t *dip)
4829 {
4830 	PMD_FUNC(pmf, "start")
4831 	int ret;
4832 	dev_info_t *pdip = ddi_get_parent(dip);
4833 	int e_pm_manage(dev_info_t *, int);
4834 	void pm_noinvol_specd(dev_info_t *dip);
4835 
4836 	e_pm_props(dip);
4837 	pm_noinvol_specd(dip);
4838 	/*
4839 	 * If this dip has already been processed, don't mess with it
4840 	 * (but decrement the speculative count we did above, as whatever
4841 	 * code put it under pm already will have dealt with it)
4842 	 */
4843 	if (PM_GET_PM_INFO(dip)) {
4844 		PMD(PMD_KIDSUP, ("%s: pm already done for %s@%s(%s#%d)\n",
4845 		    pmf, PM_DEVICE(dip)))
4846 		return (0);
4847 	}
4848 	ret = e_pm_manage(dip, PM_STYLE_UNKNOWN);
4849 
4850 	if (PM_GET_PM_INFO(dip) == NULL) {
4851 		/*
4852 		 * keep the kidsupcount increment as is
4853 		 */
4854 		DEVI(dip)->devi_pm_flags |= PMC_NOPMKID;
4855 		if (pdip && !PM_WANTS_NOTIFICATION(pdip)) {
4856 			pm_hold_power(pdip);
4857 		} else if (pdip && MDI_VHCI(pdip) && MDI_CLIENT(dip)) {
4858 			(void) mdi_power(pdip, MDI_PM_HOLD_POWER,
4859 			    (void *)dip, NULL, 0);
4860 		}
4861 
4862 		PMD(PMD_KIDSUP, ("%s: pm of %s@%s(%s#%d) failed, parent "
4863 		    "left up\n", pmf, PM_DEVICE(dip)))
4864 	}
4865 
4866 	return (ret);
4867 }
4868 
4869 /*
4870  * Keep a list of recorded thresholds.  For now we just keep a list and
4871  * search it linearly.  We don't expect too many entries.  Can always hash it
4872  * later if we need to.
4873  */
4874 void
4875 pm_record_thresh(pm_thresh_rec_t *rp)
4876 {
4877 	pm_thresh_rec_t *pptr, *ptr;
4878 
4879 	ASSERT(*rp->ptr_physpath);
4880 	rw_enter(&pm_thresh_rwlock, RW_WRITER);
4881 	for (pptr = NULL, ptr = pm_thresh_head;
4882 	    ptr; pptr = ptr,  ptr = ptr->ptr_next) {
4883 		if (strcmp(rp->ptr_physpath, ptr->ptr_physpath) == 0) {
4884 			/* replace this one */
4885 			rp->ptr_next = ptr->ptr_next;
4886 			if (pptr) {
4887 				pptr->ptr_next = rp;
4888 			} else {
4889 				pm_thresh_head = rp;
4890 			}
4891 			rw_exit(&pm_thresh_rwlock);
4892 			kmem_free(ptr, ptr->ptr_size);
4893 			return;
4894 		}
4895 		continue;
4896 	}
4897 	/*
4898 	 * There was not a match in the list, insert this one in front
4899 	 */
4900 	if (pm_thresh_head) {
4901 		rp->ptr_next = pm_thresh_head;
4902 		pm_thresh_head = rp;
4903 	} else {
4904 		rp->ptr_next = NULL;
4905 		pm_thresh_head = rp;
4906 	}
4907 	rw_exit(&pm_thresh_rwlock);
4908 }
4909 
4910 /*
4911  * Create a new dependency record and hang a new dependency entry off of it
4912  */
4913 pm_pdr_t *
4914 newpdr(char *kept, char *keeps, int isprop)
4915 {
4916 	size_t size = strlen(kept) + strlen(keeps) + 2 + sizeof (pm_pdr_t);
4917 	pm_pdr_t *p = kmem_zalloc(size, KM_SLEEP);
4918 	p->pdr_size = size;
4919 	p->pdr_isprop = isprop;
4920 	p->pdr_kept_paths = NULL;
4921 	p->pdr_kept_count = 0;
4922 	p->pdr_kept = (char *)((intptr_t)p + sizeof (pm_pdr_t));
4923 	(void) strcpy(p->pdr_kept, kept);
4924 	p->pdr_keeper = (char *)((intptr_t)p->pdr_kept + strlen(kept) + 1);
4925 	(void) strcpy(p->pdr_keeper, keeps);
4926 	ASSERT((intptr_t)p->pdr_keeper + strlen(p->pdr_keeper) + 1 <=
4927 	    (intptr_t)p + size);
4928 	ASSERT((intptr_t)p->pdr_kept + strlen(p->pdr_kept) + 1 <=
4929 	    (intptr_t)p + size);
4930 	return (p);
4931 }
4932 
4933 /*
4934  * Keep a list of recorded dependencies.  We only keep the
4935  * keeper -> kept list for simplification. At this point We do not
4936  * care about whether the devices are attached or not yet,
4937  * this would be done in pm_keeper() and pm_kept().
4938  * If a PM_RESET_PM happens, then we tear down and forget the dependencies,
4939  * and it is up to the user to issue the ioctl again if they want it
4940  * (e.g. pmconfig)
4941  * Returns true if dependency already exists in the list.
4942  */
4943 int
4944 pm_record_keeper(char *kept, char *keeper, int isprop)
4945 {
4946 	PMD_FUNC(pmf, "record_keeper")
4947 	pm_pdr_t *npdr, *ppdr, *pdr;
4948 
4949 	PMD(PMD_KEEPS, ("%s: %s, %s\n", pmf, kept, keeper))
4950 	ASSERT(kept && keeper);
4951 #ifdef DEBUG
4952 	if (pm_debug & PMD_KEEPS)
4953 		prdeps("pm_record_keeper entry");
4954 #endif
4955 	for (ppdr = NULL, pdr = pm_dep_head; pdr;
4956 	    ppdr = pdr, pdr = pdr->pdr_next) {
4957 		PMD(PMD_KEEPS, ("%s: check %s, %s\n", pmf, pdr->pdr_kept,
4958 		    pdr->pdr_keeper))
4959 		if (strcmp(kept, pdr->pdr_kept) == 0 &&
4960 		    strcmp(keeper, pdr->pdr_keeper) == 0) {
4961 			PMD(PMD_KEEPS, ("%s: match\n", pmf))
4962 			return (1);
4963 		}
4964 	}
4965 	/*
4966 	 * We did not find any match, so we have to make an entry
4967 	 */
4968 	npdr = newpdr(kept, keeper, isprop);
4969 	if (ppdr) {
4970 		ASSERT(ppdr->pdr_next == NULL);
4971 		ppdr->pdr_next = npdr;
4972 	} else {
4973 		ASSERT(pm_dep_head == NULL);
4974 		pm_dep_head = npdr;
4975 	}
4976 #ifdef DEBUG
4977 	if (pm_debug & PMD_KEEPS)
4978 		prdeps("pm_record_keeper after new record");
4979 #endif
4980 	if (!isprop)
4981 		pm_unresolved_deps++;
4982 	else
4983 		pm_prop_deps++;
4984 	return (0);
4985 }
4986 
4987 /*
4988  * Look up this device in the set of devices we've seen ioctls for
4989  * to see if we are holding a threshold spec for it.  If so, make it so.
4990  * At ioctl time, we were given the physical path of the device.
4991  */
4992 int
4993 pm_thresh_specd(dev_info_t *dip)
4994 {
4995 	void pm_apply_recorded_thresh(dev_info_t *, pm_thresh_rec_t *);
4996 	char *path = 0;
4997 	char pathbuf[MAXNAMELEN];
4998 	pm_thresh_rec_t *rp;
4999 
5000 	path = ddi_pathname(dip, pathbuf);
5001 
5002 	rw_enter(&pm_thresh_rwlock, RW_READER);
5003 	for (rp = pm_thresh_head; rp; rp = rp->ptr_next) {
5004 		if (strcmp(rp->ptr_physpath, path) != 0)
5005 			continue;
5006 		pm_apply_recorded_thresh(dip, rp);
5007 		rw_exit(&pm_thresh_rwlock);
5008 		return (1);
5009 	}
5010 	rw_exit(&pm_thresh_rwlock);
5011 	return (0);
5012 }
5013 
5014 static int
5015 pm_set_keeping(dev_info_t *keeper, dev_info_t *kept)
5016 {
5017 	PMD_FUNC(pmf, "set_keeping")
5018 	pm_info_t *kept_info;
5019 	int j, up = 0, circ;
5020 	void prdeps(char *);
5021 
5022 	PMD(PMD_KEEPS, ("%s: keeper=%s@%s(%s#%d), kept=%s@%s(%s#%d)\n", pmf,
5023 	    PM_DEVICE(keeper), PM_DEVICE(kept)))
5024 #ifdef DEBUG
5025 	if (pm_debug & PMD_KEEPS)
5026 		prdeps("Before PAD\n");
5027 #endif
5028 	ASSERT(keeper != kept);
5029 	if (PM_GET_PM_INFO(keeper) == NULL) {
5030 		cmn_err(CE_CONT, "!device %s@%s(%s#%d) keeps up device "
5031 		    "%s@%s(%s#%d), but the latter is not power managed",
5032 		    PM_DEVICE(keeper), PM_DEVICE(kept));
5033 		PMD((PMD_FAIL | PMD_KEEPS), ("%s: keeper %s@%s(%s#%d) is not"
5034 		    "power managed\n", pmf, PM_DEVICE(keeper)))
5035 		return (0);
5036 	}
5037 	kept_info = PM_GET_PM_INFO(kept);
5038 	ASSERT(kept_info);
5039 	PM_LOCK_POWER(keeper, &circ);
5040 	for (j = 0; j < PM_NUMCMPTS(keeper); j++) {
5041 		if (PM_CURPOWER(keeper, j)) {
5042 			up++;
5043 			break;
5044 		}
5045 	}
5046 	if (up) {
5047 		/* Bringup and maintain a hold on the kept */
5048 		PMD(PMD_KEEPS, ("%s: place a hold on kept %s@%s(%s#%d)\n", pmf,
5049 		    PM_DEVICE(kept)))
5050 		bring_pmdep_up(kept, 1);
5051 	}
5052 	PM_UNLOCK_POWER(keeper, circ);
5053 #ifdef DEBUG
5054 	if (pm_debug & PMD_KEEPS)
5055 		prdeps("After PAD\n");
5056 #endif
5057 	return (1);
5058 }
5059 
5060 /*
5061  * Should this device keep up another device?
5062  * Look up this device in the set of devices we've seen ioctls for
5063  * to see if we are holding a dependency spec for it.  If so, make it so.
5064  * Because we require the kept device to be attached already in order to
5065  * make the list entry (and hold it), we only need to look for keepers.
5066  * At ioctl time, we were given the physical path of the device.
5067  */
5068 int
5069 pm_keeper(char *keeper)
5070 {
5071 	PMD_FUNC(pmf, "keeper")
5072 	int pm_apply_recorded_dep(dev_info_t *, pm_pdr_t *);
5073 	dev_info_t *dip;
5074 	pm_pdr_t *dp;
5075 	dev_info_t *kept = NULL;
5076 	int ret = 0;
5077 	int i;
5078 
5079 	if (!pm_unresolved_deps && !pm_prop_deps)
5080 		return (0);
5081 	ASSERT(keeper != NULL);
5082 	dip = pm_name_to_dip(keeper, 1);
5083 	if (dip == NULL)
5084 		return (0);
5085 	PMD(PMD_KEEPS, ("%s: keeper=%s\n", pmf, keeper))
5086 	for (dp = pm_dep_head; dp; dp = dp->pdr_next) {
5087 		if (!dp->pdr_isprop) {
5088 			if (!pm_unresolved_deps)
5089 				continue;
5090 			PMD(PMD_KEEPS, ("%s: keeper %s\n", pmf, dp->pdr_keeper))
5091 			if (dp->pdr_satisfied) {
5092 				PMD(PMD_KEEPS, ("%s: satisfied\n", pmf))
5093 				continue;
5094 			}
5095 			if (strcmp(dp->pdr_keeper, keeper) == 0) {
5096 				ret += pm_apply_recorded_dep(dip, dp);
5097 			}
5098 		} else {
5099 			if (strcmp(dp->pdr_keeper, keeper) != 0)
5100 				continue;
5101 			for (i = 0; i < dp->pdr_kept_count; i++) {
5102 				if (dp->pdr_kept_paths[i] == NULL)
5103 					continue;
5104 				kept = pm_name_to_dip(dp->pdr_kept_paths[i], 1);
5105 				if (kept == NULL)
5106 					continue;
5107 				ASSERT(ddi_prop_exists(DDI_DEV_T_ANY, kept,
5108 				    DDI_PROP_DONTPASS, dp->pdr_kept));
5109 				PMD(PMD_KEEPS, ("%s: keeper=%s@%s(%s#%d), "
5110 				    "kept=%s@%s(%s#%d) keptcnt=%d\n",
5111 				    pmf, PM_DEVICE(dip), PM_DEVICE(kept),
5112 				    dp->pdr_kept_count))
5113 				if (kept != dip) {
5114 					ret += pm_set_keeping(dip, kept);
5115 				}
5116 				ddi_release_devi(kept);
5117 			}
5118 
5119 		}
5120 	}
5121 	ddi_release_devi(dip);
5122 	return (ret);
5123 }
5124 
5125 /*
5126  * Should this device be kept up by another device?
5127  * Look up all dependency recorded from PM_ADD_DEPENDENT and
5128  * PM_ADD_DEPENDENT_PROPERTY ioctls. Record down on the keeper's
5129  * kept device lists.
5130  */
5131 static int
5132 pm_kept(char *keptp)
5133 {
5134 	PMD_FUNC(pmf, "kept")
5135 	pm_pdr_t *dp;
5136 	int found = 0;
5137 	int ret = 0;
5138 	dev_info_t *keeper;
5139 	dev_info_t *kept;
5140 	size_t length;
5141 	int i;
5142 	char **paths;
5143 	char *path;
5144 
5145 	ASSERT(keptp != NULL);
5146 	kept = pm_name_to_dip(keptp, 1);
5147 	if (kept == NULL)
5148 		return (0);
5149 	PMD(PMD_KEEPS, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(kept)))
5150 	for (dp = pm_dep_head; dp; dp = dp->pdr_next) {
5151 		if (dp->pdr_isprop) {
5152 			PMD(PMD_KEEPS, ("%s: property %s\n", pmf, dp->pdr_kept))
5153 			if (ddi_prop_exists(DDI_DEV_T_ANY, kept,
5154 			    DDI_PROP_DONTPASS, dp->pdr_kept)) {
5155 				/*
5156 				 * Dont allow self dependency.
5157 				 */
5158 				if (strcmp(dp->pdr_keeper, keptp) == 0)
5159 					continue;
5160 				keeper = pm_name_to_dip(dp->pdr_keeper, 1);
5161 				if (keeper == NULL)
5162 					continue;
5163 				PMD(PMD_KEEPS, ("%s: adding to kepts path list "
5164 				    "%p\n", pmf, (void *)kept))
5165 #ifdef DEBUG
5166 				if (pm_debug & PMD_DEP)
5167 					prdeps("Before Adding from pm_kept\n");
5168 #endif
5169 				/*
5170 				 * Add ourselves to the dip list.
5171 				 */
5172 				if (dp->pdr_kept_count == 0) {
5173 					length = strlen(keptp) + 1;
5174 					path =
5175 					    kmem_alloc(length, KM_SLEEP);
5176 					paths = kmem_alloc(sizeof (char **),
5177 						    KM_SLEEP);
5178 					(void) strcpy(path, keptp);
5179 					paths[0] = path;
5180 					dp->pdr_kept_paths = paths;
5181 					dp->pdr_kept_count++;
5182 				} else {
5183 					/* Check to see if already on list */
5184 					for (i = 0; i < dp->pdr_kept_count;
5185 					    i++) {
5186 						if (strcmp(keptp,
5187 						    dp->pdr_kept_paths[i])
5188 						    == 0) {
5189 							found++;
5190 							break;
5191 						}
5192 					}
5193 					if (found) {
5194 						ddi_release_devi(keeper);
5195 						continue;
5196 					}
5197 					length = dp->pdr_kept_count *
5198 					    sizeof (char **);
5199 					paths = kmem_alloc(
5200 					    length + sizeof (char **),
5201 					    KM_SLEEP);
5202 					if (dp->pdr_kept_count) {
5203 						bcopy(dp->pdr_kept_paths,
5204 						    paths, length);
5205 						kmem_free(dp->pdr_kept_paths,
5206 							length);
5207 					}
5208 					dp->pdr_kept_paths = paths;
5209 					length = strlen(keptp) + 1;
5210 					path =
5211 					    kmem_alloc(length, KM_SLEEP);
5212 					(void) strcpy(path, keptp);
5213 					dp->pdr_kept_paths[i] = path;
5214 					dp->pdr_kept_count++;
5215 				}
5216 #ifdef DEBUG
5217 				if (pm_debug & PMD_DEP)
5218 					prdeps("After from pm_kept\n");
5219 #endif
5220 				if (keeper) {
5221 					ret += pm_set_keeping(keeper, kept);
5222 					ddi_release_devi(keeper);
5223 				}
5224 			}
5225 		} else {
5226 			/*
5227 			 * pm_keeper would be called later to do
5228 			 * the actual pm_set_keeping.
5229 			 */
5230 			PMD(PMD_KEEPS, ("%s: adding to kepts path list %p\n",
5231 			    pmf, (void *)kept))
5232 #ifdef DEBUG
5233 			if (pm_debug & PMD_DEP)
5234 				prdeps("Before Adding from pm_kept\n");
5235 #endif
5236 			if (strcmp(keptp, dp->pdr_kept) == 0) {
5237 				if (dp->pdr_kept_paths == NULL) {
5238 					length = strlen(keptp) + 1;
5239 					path =
5240 					    kmem_alloc(length, KM_SLEEP);
5241 					paths = kmem_alloc(sizeof (char **),
5242 						KM_SLEEP);
5243 					(void) strcpy(path, keptp);
5244 					paths[0] = path;
5245 					dp->pdr_kept_paths = paths;
5246 					dp->pdr_kept_count++;
5247 				}
5248 			}
5249 #ifdef DEBUG
5250 			if (pm_debug & PMD_DEP)
5251 			    prdeps("After from pm_kept\n");
5252 #endif
5253 		}
5254 	}
5255 	ddi_release_devi(kept);
5256 	return (ret);
5257 }
5258 
5259 /*
5260  * Apply a recorded dependency.  dp specifies the dependency, and
5261  * keeper is already known to be the device that keeps up the other (kept) one.
5262  * We have to the whole tree for the "kept" device, then apply
5263  * the dependency (which may already be applied).
5264  */
5265 int
5266 pm_apply_recorded_dep(dev_info_t *keeper, pm_pdr_t *dp)
5267 {
5268 	PMD_FUNC(pmf, "apply_recorded_dep")
5269 	dev_info_t *kept = NULL;
5270 	int ret = 0;
5271 	char *keptp = NULL;
5272 
5273 	/*
5274 	 * Device to Device dependency can only be 1 to 1.
5275 	 */
5276 	if (dp->pdr_kept_paths == NULL)
5277 		return (0);
5278 	keptp = dp->pdr_kept_paths[0];
5279 	if (keptp == NULL)
5280 		return (0);
5281 	ASSERT(*keptp != '\0');
5282 	kept = pm_name_to_dip(keptp, 1);
5283 	if (kept == NULL)
5284 		return (0);
5285 	if (kept) {
5286 		PMD(PMD_KEEPS, ("%s: keeper=%s, kept=%s\n", pmf,
5287 		    dp->pdr_keeper, keptp))
5288 		if (pm_set_keeping(keeper, kept)) {
5289 			ASSERT(dp->pdr_satisfied == 0);
5290 			dp->pdr_satisfied = 1;
5291 			ASSERT(pm_unresolved_deps);
5292 			pm_unresolved_deps--;
5293 			ret++;
5294 		}
5295 	}
5296 	ddi_release_devi(kept);
5297 
5298 	return (ret);
5299 }
5300 
5301 /*
5302  * Called from common/io/pm.c
5303  */
5304 int
5305 pm_cur_power(pm_component_t *cp)
5306 {
5307 	return (cur_power(cp));
5308 }
5309 
5310 /*
5311  * External interface to sanity-check a power level.
5312  */
5313 int
5314 pm_valid_power(dev_info_t *dip, int comp, int level)
5315 {
5316 	PMD_FUNC(pmf, "valid_power")
5317 
5318 	if (comp >= 0 && comp < PM_NUMCMPTS(dip) && level >= 0)
5319 		return (e_pm_valid_power(dip, comp, level));
5320 	else {
5321 		PMD(PMD_FAIL, ("%s: comp=%d, ncomp=%d, level=%d\n",
5322 		    pmf, comp, PM_NUMCMPTS(dip), level))
5323 		return (0);
5324 	}
5325 }
5326 
5327 /*
5328  * Called when a device that is direct power managed needs to change state.
5329  * This routine arranges to block the request until the process managing
5330  * the device makes the change (or some other incompatible change) or
5331  * the process closes /dev/pm.
5332  */
5333 static int
5334 pm_block(dev_info_t *dip, int comp, int newpower, int oldpower)
5335 {
5336 	pm_rsvp_t *new = kmem_zalloc(sizeof (*new), KM_SLEEP);
5337 	int ret = 0;
5338 	void pm_dequeue_blocked(pm_rsvp_t *);
5339 	void pm_enqueue_blocked(pm_rsvp_t *);
5340 
5341 	ASSERT(!pm_processes_stopped);
5342 	ASSERT(PM_IAM_LOCKING_DIP(dip));
5343 	new->pr_dip = dip;
5344 	new->pr_comp = comp;
5345 	new->pr_newlevel = newpower;
5346 	new->pr_oldlevel = oldpower;
5347 	cv_init(&new->pr_cv, NULL, CV_DEFAULT, NULL);
5348 	mutex_enter(&pm_rsvp_lock);
5349 	pm_enqueue_blocked(new);
5350 	pm_enqueue_notify(PSC_PENDING_CHANGE, dip, comp, newpower, oldpower,
5351 	    PM_CANBLOCK_BLOCK);
5352 	PM_UNLOCK_DIP(dip);
5353 	/*
5354 	 * truss may make the cv_wait_sig return prematurely
5355 	 */
5356 	while (ret == 0) {
5357 		/*
5358 		 * Normally there will be no user context involved, but if
5359 		 * there is (e.g. we are here via an ioctl call to a driver)
5360 		 * then we should allow the process to abort the request,
5361 		 * or we get an unkillable process if the same thread does
5362 		 * PM_DIRECT_PM and pm_raise_power
5363 		 */
5364 		if (cv_wait_sig(&new->pr_cv, &pm_rsvp_lock) == 0) {
5365 			ret = PMP_FAIL;
5366 		} else {
5367 			ret = new->pr_retval;
5368 		}
5369 	}
5370 	pm_dequeue_blocked(new);
5371 	mutex_exit(&pm_rsvp_lock);
5372 	cv_destroy(&new->pr_cv);
5373 	kmem_free(new, sizeof (*new));
5374 	return (ret);
5375 }
5376 
5377 /*
5378  * Returns true if the process is interested in power level changes (has issued
5379  * PM_GET_STATE_CHANGE ioctl).
5380  */
5381 int
5382 pm_interest_registered(int clone)
5383 {
5384 	ASSERT(clone >= 0 && clone < PM_MAX_CLONE - 1);
5385 	return (pm_interest[clone]);
5386 }
5387 
5388 /*
5389  * Process with clone has just done PM_DIRECT_PM on dip, or has asked to
5390  * watch all state transitions (dip == NULL).  Set up data
5391  * structs to communicate with process about state changes.
5392  */
5393 void
5394 pm_register_watcher(int clone, dev_info_t *dip)
5395 {
5396 	pscc_t	*p;
5397 	psce_t	*psce;
5398 	static void pm_enqueue_pscc(pscc_t *, pscc_t **);
5399 
5400 	/*
5401 	 * We definitely need a control struct, then we have to search to see
5402 	 * there is already an entries struct (in the dip != NULL case).
5403 	 */
5404 	pscc_t	*pscc = kmem_zalloc(sizeof (*pscc), KM_SLEEP);
5405 	pscc->pscc_clone = clone;
5406 	pscc->pscc_dip = dip;
5407 
5408 	if (dip) {
5409 		int found = 0;
5410 		rw_enter(&pm_pscc_direct_rwlock, RW_WRITER);
5411 		for (p = pm_pscc_direct; p; p = p->pscc_next) {
5412 			/*
5413 			 * Already an entry for this clone, so just use it
5414 			 * for the new one (for the case where a single
5415 			 * process is watching multiple devices)
5416 			 */
5417 			if (p->pscc_clone == clone) {
5418 				ASSERT(p->pscc_dip != dip);
5419 				pscc->pscc_entries = p->pscc_entries;
5420 				pscc->pscc_entries->psce_references++;
5421 				found++;
5422 			}
5423 		}
5424 		if (!found) {		/* create a new one */
5425 			psce = kmem_zalloc(sizeof (psce_t), KM_SLEEP);
5426 			mutex_init(&psce->psce_lock, NULL, MUTEX_DEFAULT, NULL);
5427 			psce->psce_first =
5428 			    kmem_zalloc(sizeof (pm_state_change_t) * PSCCOUNT,
5429 			    KM_SLEEP);
5430 			psce->psce_in = psce->psce_out = psce->psce_first;
5431 			psce->psce_last = &psce->psce_first[PSCCOUNT - 1];
5432 			psce->psce_references = 1;
5433 			pscc->pscc_entries = psce;
5434 		}
5435 		pm_enqueue_pscc(pscc, &pm_pscc_direct);
5436 		rw_exit(&pm_pscc_direct_rwlock);
5437 	} else {
5438 		ASSERT(!pm_interest_registered(clone));
5439 		rw_enter(&pm_pscc_interest_rwlock, RW_WRITER);
5440 #ifdef DEBUG
5441 		for (p = pm_pscc_interest; p; p = p->pscc_next) {
5442 			/*
5443 			 * Should not be an entry for this clone!
5444 			 */
5445 			ASSERT(p->pscc_clone != clone);
5446 		}
5447 #endif
5448 		psce = kmem_zalloc(sizeof (psce_t), KM_SLEEP);
5449 		psce->psce_first = kmem_zalloc(sizeof (pm_state_change_t) *
5450 		    PSCCOUNT, KM_SLEEP);
5451 		psce->psce_in = psce->psce_out = psce->psce_first;
5452 		psce->psce_last = &psce->psce_first[PSCCOUNT - 1];
5453 		psce->psce_references = 1;
5454 		pscc->pscc_entries = psce;
5455 		pm_enqueue_pscc(pscc, &pm_pscc_interest);
5456 		pm_interest[clone] = 1;
5457 		rw_exit(&pm_pscc_interest_rwlock);
5458 	}
5459 }
5460 
5461 /*
5462  * Remove the given entry from the blocked list
5463  */
5464 void
5465 pm_dequeue_blocked(pm_rsvp_t *p)
5466 {
5467 	ASSERT(MUTEX_HELD(&pm_rsvp_lock));
5468 	if (pm_blocked_list == p) {
5469 		ASSERT(p->pr_prev == NULL);
5470 		if (p->pr_next != NULL)
5471 			p->pr_next->pr_prev = NULL;
5472 		pm_blocked_list = p->pr_next;
5473 	} else {
5474 		ASSERT(p->pr_prev != NULL);
5475 		p->pr_prev->pr_next = p->pr_next;
5476 		if (p->pr_next != NULL)
5477 			p->pr_next->pr_prev = p->pr_prev;
5478 	}
5479 }
5480 
5481 /*
5482  * Remove the given control struct from the given list
5483  */
5484 static void
5485 pm_dequeue_pscc(pscc_t *p, pscc_t **list)
5486 {
5487 	if (*list == p) {
5488 		ASSERT(p->pscc_prev == NULL);
5489 		if (p->pscc_next != NULL)
5490 			p->pscc_next->pscc_prev = NULL;
5491 		*list = p->pscc_next;
5492 	} else {
5493 		ASSERT(p->pscc_prev != NULL);
5494 		p->pscc_prev->pscc_next = p->pscc_next;
5495 		if (p->pscc_next != NULL)
5496 			p->pscc_next->pscc_prev = p->pscc_prev;
5497 	}
5498 }
5499 
5500 /*
5501  * Stick the control struct specified on the front of the list
5502  */
5503 static void
5504 pm_enqueue_pscc(pscc_t *p, pscc_t **list)
5505 {
5506 	pscc_t *h;	/* entry at head of list */
5507 	if ((h = *list) == NULL) {
5508 		*list = p;
5509 		ASSERT(p->pscc_next == NULL);
5510 		ASSERT(p->pscc_prev == NULL);
5511 	} else {
5512 		p->pscc_next = h;
5513 		ASSERT(h->pscc_prev == NULL);
5514 		h->pscc_prev = p;
5515 		ASSERT(p->pscc_prev == NULL);
5516 		*list = p;
5517 	}
5518 }
5519 
5520 /*
5521  * If dip is NULL, process is closing "clone" clean up all its registrations.
5522  * Otherwise only clean up those for dip because process is just giving up
5523  * control of a direct device.
5524  */
5525 void
5526 pm_deregister_watcher(int clone, dev_info_t *dip)
5527 {
5528 	pscc_t	*p, *pn;
5529 	psce_t	*psce;
5530 	int found = 0;
5531 
5532 	if (dip == NULL) {
5533 		rw_enter(&pm_pscc_interest_rwlock, RW_WRITER);
5534 		for (p = pm_pscc_interest; p; p = pn) {
5535 			pn = p->pscc_next;
5536 			if (p->pscc_clone == clone) {
5537 				pm_dequeue_pscc(p, &pm_pscc_interest);
5538 				psce = p->pscc_entries;
5539 				ASSERT(psce->psce_references == 1);
5540 				mutex_destroy(&psce->psce_lock);
5541 				kmem_free(psce->psce_first,
5542 				    sizeof (pm_state_change_t) * PSCCOUNT);
5543 				kmem_free(psce, sizeof (*psce));
5544 				kmem_free(p, sizeof (*p));
5545 			}
5546 		}
5547 		pm_interest[clone] = 0;
5548 		rw_exit(&pm_pscc_interest_rwlock);
5549 	}
5550 	found = 0;
5551 	rw_enter(&pm_pscc_direct_rwlock, RW_WRITER);
5552 	for (p = pm_pscc_direct; p; p = pn) {
5553 		pn = p->pscc_next;
5554 		if ((dip && p->pscc_dip == dip) ||
5555 		    (dip == NULL && clone == p->pscc_clone)) {
5556 			ASSERT(clone == p->pscc_clone);
5557 			found++;
5558 			/*
5559 			 * Remove from control list
5560 			 */
5561 			pm_dequeue_pscc(p, &pm_pscc_direct);
5562 			/*
5563 			 * If we're the last reference, free the
5564 			 * entries struct.
5565 			 */
5566 			psce = p->pscc_entries;
5567 			ASSERT(psce);
5568 			if (psce->psce_references == 1) {
5569 				kmem_free(psce->psce_first,
5570 				    PSCCOUNT * sizeof (pm_state_change_t));
5571 				kmem_free(psce, sizeof (*psce));
5572 			} else {
5573 				psce->psce_references--;
5574 			}
5575 			kmem_free(p, sizeof (*p));
5576 		}
5577 	}
5578 	ASSERT(dip == NULL || found);
5579 	rw_exit(&pm_pscc_direct_rwlock);
5580 }
5581 
5582 /*
5583  * Search the indicated list for an entry that matches clone, and return a
5584  * pointer to it.  To be interesting, the entry must have something ready to
5585  * be passed up to the controlling process.
5586  * The returned entry will be locked upon return from this call.
5587  */
5588 static psce_t *
5589 pm_psc_find_clone(int clone, pscc_t **list, krwlock_t *lock)
5590 {
5591 	pscc_t	*p;
5592 	psce_t	*psce;
5593 	rw_enter(lock, RW_READER);
5594 	for (p = *list; p; p = p->pscc_next) {
5595 		if (clone == p->pscc_clone) {
5596 			psce = p->pscc_entries;
5597 			mutex_enter(&psce->psce_lock);
5598 			if (psce->psce_out->size) {
5599 				rw_exit(lock);
5600 				return (psce);
5601 			} else {
5602 				mutex_exit(&psce->psce_lock);
5603 			}
5604 		}
5605 	}
5606 	rw_exit(lock);
5607 	return (NULL);
5608 }
5609 
5610 /*
5611  * Find an entry for a particular clone in the direct list.
5612  */
5613 psce_t *
5614 pm_psc_clone_to_direct(int clone)
5615 {
5616 	static psce_t *pm_psc_find_clone(int, pscc_t **, krwlock_t *);
5617 	return (pm_psc_find_clone(clone, &pm_pscc_direct,
5618 	    &pm_pscc_direct_rwlock));
5619 }
5620 
5621 /*
5622  * Find an entry for a particular clone in the interest list.
5623  */
5624 psce_t *
5625 pm_psc_clone_to_interest(int clone)
5626 {
5627 	static psce_t *pm_psc_find_clone(int, pscc_t **, krwlock_t *);
5628 	return (pm_psc_find_clone(clone, &pm_pscc_interest,
5629 	    &pm_pscc_interest_rwlock));
5630 }
5631 
5632 /*
5633  * Put the given entry at the head of the blocked list
5634  */
5635 void
5636 pm_enqueue_blocked(pm_rsvp_t *p)
5637 {
5638 	ASSERT(MUTEX_HELD(&pm_rsvp_lock));
5639 	ASSERT(p->pr_next == NULL);
5640 	ASSERT(p->pr_prev == NULL);
5641 	if (pm_blocked_list != NULL) {
5642 		p->pr_next = pm_blocked_list;
5643 		ASSERT(pm_blocked_list->pr_prev == NULL);
5644 		pm_blocked_list->pr_prev = p;
5645 		pm_blocked_list = p;
5646 	} else {
5647 		pm_blocked_list = p;
5648 	}
5649 }
5650 
5651 /*
5652  * Sets every power managed device back to its default threshold
5653  */
5654 void
5655 pm_all_to_default_thresholds(void)
5656 {
5657 	ddi_walk_devs(ddi_root_node(), pm_set_dev_thr_walk,
5658 	    (void *) &pm_system_idle_threshold);
5659 }
5660 
5661 static int
5662 pm_set_dev_thr_walk(dev_info_t *dip, void *arg)
5663 {
5664 	int thr = (int)(*(int *)arg);
5665 
5666 	if (!PM_GET_PM_INFO(dip))
5667 		return (DDI_WALK_CONTINUE);
5668 	pm_set_device_threshold(dip, thr, PMC_DEF_THRESH);
5669 	return (DDI_WALK_CONTINUE);
5670 }
5671 
5672 /*
5673  * Returns the current threshold value (in seconds) for the indicated component
5674  */
5675 int
5676 pm_current_threshold(dev_info_t *dip, int comp, int *threshp)
5677 {
5678 	if (comp < 0 || comp >= PM_NUMCMPTS(dip)) {
5679 		return (DDI_FAILURE);
5680 	} else {
5681 		*threshp = cur_threshold(dip, comp);
5682 		return (DDI_SUCCESS);
5683 	}
5684 }
5685 
5686 /*
5687  * To be called when changing the power level of a component of a device.
5688  * On some platforms, changing power on one device may require that power
5689  * be changed on other, related devices in the same transaction.  Thus, we
5690  * always pass this request to the platform power manager so that all the
5691  * affected devices will be locked.
5692  */
5693 void
5694 pm_lock_power(dev_info_t *dip, int *circp)
5695 {
5696 	power_req_t power_req;
5697 	int result;
5698 
5699 	power_req.request_type = PMR_PPM_LOCK_POWER;
5700 	power_req.req.ppm_lock_power_req.who = dip;
5701 	power_req.req.ppm_lock_power_req.circp = circp;
5702 	(void) pm_ctlops(PPM(dip), dip, DDI_CTLOPS_POWER, &power_req, &result);
5703 }
5704 
5705 /*
5706  * Release the lock (or locks) acquired to change the power of a device.
5707  * See comments for pm_lock_power.
5708  */
5709 void
5710 pm_unlock_power(dev_info_t *dip, int circ)
5711 {
5712 	power_req_t power_req;
5713 	int result;
5714 
5715 	power_req.request_type = PMR_PPM_UNLOCK_POWER;
5716 	power_req.req.ppm_unlock_power_req.who = dip;
5717 	power_req.req.ppm_unlock_power_req.circ = circ;
5718 	(void) pm_ctlops(PPM(dip), dip, DDI_CTLOPS_POWER, &power_req, &result);
5719 }
5720 
5721 
5722 /*
5723  * Attempt (without blocking) to acquire the lock(s) needed to change the
5724  * power of a component of a device.  See comments for pm_lock_power.
5725  *
5726  * Return: 1 if lock(s) acquired, 0 if not.
5727  */
5728 int
5729 pm_try_locking_power(dev_info_t *dip, int *circp)
5730 {
5731 	power_req_t power_req;
5732 	int result;
5733 
5734 	power_req.request_type = PMR_PPM_TRY_LOCK_POWER;
5735 	power_req.req.ppm_lock_power_req.who = dip;
5736 	power_req.req.ppm_lock_power_req.circp = circp;
5737 	(void) pm_ctlops(PPM(dip), dip, DDI_CTLOPS_POWER, &power_req, &result);
5738 	return (result);
5739 }
5740 
5741 
5742 /*
5743  * Lock power state of a device.
5744  *
5745  * The implementation handles a special case where another thread may have
5746  * acquired the lock and created/launched this thread to do the work.  If
5747  * the lock cannot be acquired immediately, we check to see if this thread
5748  * is registered as a borrower of the lock.  If so, we may proceed without
5749  * the lock.  This assumes that the lending thread blocks on the completion
5750  * of this thread.
5751  *
5752  * Note 1: for use by ppm only.
5753  *
5754  * Note 2: On failing to get the lock immediately, we search lock_loan list
5755  * for curthread (as borrower of the lock).  On a hit, we check that the
5756  * lending thread already owns the lock we want.  It is safe to compare
5757  * devi_busy_thread and thread id of the lender because in the == case (the
5758  * only one we care about) we know that the owner is blocked.  Similarly,
5759  * If we find that curthread isn't registered as a lock borrower, it is safe
5760  * to use the blocking call (ndi_devi_enter) because we know that if we
5761  * weren't already listed as a borrower (upstream on the call stack) we won't
5762  * become one.
5763  */
5764 void
5765 pm_lock_power_single(dev_info_t *dip, int *circp)
5766 {
5767 	lock_loan_t *cur;
5768 
5769 	/* if the lock is available, we are done. */
5770 	if (ndi_devi_tryenter(dip, circp))
5771 		return;
5772 
5773 	mutex_enter(&pm_loan_lock);
5774 	/* see if our thread is registered as a lock borrower. */
5775 	for (cur = lock_loan_head.pmlk_next; cur; cur = cur->pmlk_next)
5776 		if (cur->pmlk_borrower == curthread)
5777 			break;
5778 	mutex_exit(&pm_loan_lock);
5779 
5780 	/* if this thread not already registered, it is safe to block */
5781 	if (cur == NULL)
5782 		ndi_devi_enter(dip, circp);
5783 	else {
5784 		/* registered: does lender own the lock we want? */
5785 		if (cur->pmlk_lender == DEVI(dip)->devi_busy_thread) {
5786 			ASSERT(cur->pmlk_dip == NULL || cur->pmlk_dip == dip);
5787 			cur->pmlk_dip = dip;
5788 		} else /* no: just block for it */
5789 			ndi_devi_enter(dip, circp);
5790 
5791 	}
5792 }
5793 
5794 /*
5795  * Drop the lock on the device's power state.  See comment for
5796  * pm_lock_power_single() for special implementation considerations.
5797  *
5798  * Note: for use by ppm only.
5799  */
5800 void
5801 pm_unlock_power_single(dev_info_t *dip, int circ)
5802 {
5803 	lock_loan_t *cur;
5804 
5805 	/* optimization: mutex not needed to check empty list */
5806 	if (lock_loan_head.pmlk_next == NULL) {
5807 		ndi_devi_exit(dip, circ);
5808 		return;
5809 	}
5810 
5811 	mutex_enter(&pm_loan_lock);
5812 	/* see if our thread is registered as a lock borrower. */
5813 	for (cur = lock_loan_head.pmlk_next; cur; cur = cur->pmlk_next)
5814 		if (cur->pmlk_borrower == curthread)
5815 			break;
5816 	mutex_exit(&pm_loan_lock);
5817 
5818 	if (cur == NULL || cur->pmlk_dip != dip)
5819 		/* we acquired the lock directly, so return it */
5820 		ndi_devi_exit(dip, circ);
5821 }
5822 
5823 /*
5824  * Try to take the lock for changing the power level of a component.
5825  *
5826  * Note: for use by ppm only.
5827  */
5828 int
5829 pm_try_locking_power_single(dev_info_t *dip, int *circp)
5830 {
5831 	return (ndi_devi_tryenter(dip, circp));
5832 }
5833 
5834 #ifdef	DEBUG
5835 /*
5836  * The following are used only to print out data structures for debugging
5837  */
5838 void
5839 prdeps(char *msg)
5840 {
5841 
5842 	pm_pdr_t *rp;
5843 	int i;
5844 
5845 	pm_log("pm_dep_head %s %p\n", msg, (void *)pm_dep_head);
5846 	for (rp = pm_dep_head; rp; rp = rp->pdr_next) {
5847 		pm_log("%p: %s keeper %s, kept %s, kept count %d, next %p\n",
5848 		    (void *)rp, (rp->pdr_isprop ? "property" : "device"),
5849 		    rp->pdr_keeper, rp->pdr_kept, rp->pdr_kept_count,
5850 		    (void *)rp->pdr_next);
5851 		if (rp->pdr_kept_count != 0) {
5852 			pm_log("kept list = ");
5853 			i = 0;
5854 			while (i < rp->pdr_kept_count) {
5855 				pm_log("%s ", rp->pdr_kept_paths[i]);
5856 				i++;
5857 			}
5858 			pm_log("\n");
5859 		}
5860 	}
5861 }
5862 
5863 void
5864 pr_noinvol(char *hdr)
5865 {
5866 	pm_noinvol_t *ip;
5867 
5868 	pm_log("%s\n", hdr);
5869 	rw_enter(&pm_noinvol_rwlock, RW_READER);
5870 	for (ip = pm_noinvol_head; ip; ip = ip->ni_next)
5871 		pm_log("\tmaj %d, flags %x, noinvolpm %d %s\n",
5872 		    ip->ni_major, ip->ni_flags, ip->ni_noinvolpm, ip->ni_path);
5873 	rw_exit(&pm_noinvol_rwlock);
5874 }
5875 #endif
5876 
5877 /*
5878  * Attempt to apply the thresholds indicated by rp to the node specified by
5879  * dip.
5880  */
5881 void
5882 pm_apply_recorded_thresh(dev_info_t *dip, pm_thresh_rec_t *rp)
5883 {
5884 	PMD_FUNC(pmf, "apply_recorded_thresh")
5885 	int i, j;
5886 	int comps = PM_NUMCMPTS(dip);
5887 	struct pm_component *cp;
5888 	pm_pte_t *ep;
5889 	int pm_valid_thresh(dev_info_t *, pm_thresh_rec_t *);
5890 
5891 	PMD(PMD_THRESH, ("%s: part: %s@%s(%s#%d), rp %p, %s\n", pmf,
5892 	    PM_DEVICE(dip), (void *)rp, rp->ptr_physpath))
5893 	PM_LOCK_DIP(dip);
5894 	if (!PM_GET_PM_INFO(dip) || PM_ISBC(dip) || !pm_valid_thresh(dip, rp)) {
5895 		PMD(PMD_FAIL, ("%s: part: %s@%s(%s#%d) PM_GET_PM_INFO %p\n",
5896 		    pmf, PM_DEVICE(dip), (void*)PM_GET_PM_INFO(dip)))
5897 		PMD(PMD_FAIL, ("%s: part: %s@%s(%s#%d) PM_ISBC %d\n",
5898 		    pmf, PM_DEVICE(dip), PM_ISBC(dip)))
5899 		PMD(PMD_FAIL, ("%s: part: %s@%s(%s#%d) pm_valid_thresh %d\n",
5900 		    pmf, PM_DEVICE(dip), pm_valid_thresh(dip, rp)))
5901 		PM_UNLOCK_DIP(dip);
5902 		return;
5903 	}
5904 
5905 	ep = rp->ptr_entries;
5906 	/*
5907 	 * Here we do the special case of a device threshold
5908 	 */
5909 	if (rp->ptr_numcomps == 0) {	/* PM_SET_DEVICE_THRESHOLD product */
5910 		ASSERT(ep && ep->pte_numthresh == 1);
5911 		PMD(PMD_THRESH, ("%s: set dev thr %s@%s(%s#%d) to 0x%x\n",
5912 		    pmf, PM_DEVICE(dip), ep->pte_thresh[0]))
5913 		PM_UNLOCK_DIP(dip);
5914 		pm_set_device_threshold(dip, ep->pte_thresh[0], PMC_DEV_THRESH);
5915 		if (autopm_enabled)
5916 			pm_rescan(dip);
5917 		return;
5918 	}
5919 	for (i = 0; i < comps; i++) {
5920 		cp = PM_CP(dip, i);
5921 		for (j = 0; j < ep->pte_numthresh; j++) {
5922 			PMD(PMD_THRESH, ("%s: set thr %d for %s@%s(%s#%d)[%d] "
5923 			    "to %x\n", pmf, j, PM_DEVICE(dip),
5924 			    i, ep->pte_thresh[j]))
5925 			cp->pmc_comp.pmc_thresh[j + 1] = ep->pte_thresh[j];
5926 		}
5927 		ep++;
5928 	}
5929 	DEVI(dip)->devi_pm_flags &= PMC_THRESH_NONE;
5930 	DEVI(dip)->devi_pm_flags |= PMC_COMP_THRESH;
5931 	PM_UNLOCK_DIP(dip);
5932 
5933 	if (autopm_enabled)
5934 		pm_rescan(dip);
5935 }
5936 
5937 /*
5938  * Returns true if the threshold specified by rp could be applied to dip
5939  * (that is, the number of components and transitions are the same)
5940  */
5941 int
5942 pm_valid_thresh(dev_info_t *dip, pm_thresh_rec_t *rp)
5943 {
5944 	PMD_FUNC(pmf, "valid_thresh")
5945 	int comps, i;
5946 	pm_component_t *cp;
5947 	pm_pte_t *ep;
5948 
5949 	if (!PM_GET_PM_INFO(dip) || PM_ISBC(dip)) {
5950 		PMD(PMD_ERROR, ("%s: %s: no pm_info or BC\n", pmf,
5951 		    rp->ptr_physpath))
5952 		return (0);
5953 	}
5954 	/*
5955 	 * Special case: we represent the PM_SET_DEVICE_THRESHOLD case by
5956 	 * an entry with numcomps == 0, (since we don't know how many
5957 	 * components there are in advance).  This is always a valid
5958 	 * spec.
5959 	 */
5960 	if (rp->ptr_numcomps == 0) {
5961 		ASSERT(rp->ptr_entries && rp->ptr_entries->pte_numthresh == 1);
5962 		return (1);
5963 	}
5964 	if (rp->ptr_numcomps != (comps = PM_NUMCMPTS(dip))) {
5965 		PMD(PMD_ERROR, ("%s: comp # mm (dip %d cmd %d) for %s\n",
5966 		    pmf, PM_NUMCMPTS(dip), rp->ptr_numcomps, rp->ptr_physpath))
5967 		return (0);
5968 	}
5969 	ep = rp->ptr_entries;
5970 	for (i = 0; i < comps; i++) {
5971 		cp = PM_CP(dip, i);
5972 		if ((ep + i)->pte_numthresh !=
5973 		    cp->pmc_comp.pmc_numlevels - 1) {
5974 			PMD(PMD_ERROR, ("%s: %s[%d]: thresh=%d, record=%d\n",
5975 			    pmf, rp->ptr_physpath, i,
5976 			    cp->pmc_comp.pmc_numlevels - 1,
5977 			    (ep + i)->pte_numthresh))
5978 			return (0);
5979 		}
5980 	}
5981 	return (1);
5982 }
5983 
5984 /*
5985  * Remove any recorded threshold for device physpath
5986  * We know there will be at most one.
5987  */
5988 void
5989 pm_unrecord_threshold(char *physpath)
5990 {
5991 	pm_thresh_rec_t *pptr, *ptr;
5992 
5993 	rw_enter(&pm_thresh_rwlock, RW_WRITER);
5994 	for (pptr = NULL, ptr = pm_thresh_head; ptr; ptr = ptr->ptr_next) {
5995 		if (strcmp(physpath, ptr->ptr_physpath) == 0) {
5996 			if (pptr) {
5997 				pptr->ptr_next = ptr->ptr_next;
5998 			} else {
5999 				ASSERT(pm_thresh_head == ptr);
6000 				pm_thresh_head = ptr->ptr_next;
6001 			}
6002 			kmem_free(ptr, ptr->ptr_size);
6003 			break;
6004 		}
6005 		pptr = ptr;
6006 	}
6007 	rw_exit(&pm_thresh_rwlock);
6008 }
6009 
6010 /*
6011  * Discard all recorded thresholds.  We are returning to the default pm state.
6012  */
6013 void
6014 pm_discard_thresholds(void)
6015 {
6016 	pm_thresh_rec_t *rp;
6017 	rw_enter(&pm_thresh_rwlock, RW_WRITER);
6018 	while (pm_thresh_head) {
6019 		rp = pm_thresh_head;
6020 		pm_thresh_head = rp->ptr_next;
6021 		kmem_free(rp, rp->ptr_size);
6022 	}
6023 	rw_exit(&pm_thresh_rwlock);
6024 }
6025 
6026 /*
6027  * Discard all recorded dependencies.  We are returning to the default pm state.
6028  */
6029 void
6030 pm_discard_dependencies(void)
6031 {
6032 	pm_pdr_t *rp;
6033 	int i;
6034 	size_t length;
6035 
6036 #ifdef DEBUG
6037 	if (pm_debug & PMD_DEP)
6038 		prdeps("Before discard\n");
6039 #endif
6040 	ddi_walk_devs(ddi_root_node(), pm_discard_dep_walk, NULL);
6041 
6042 #ifdef DEBUG
6043 	if (pm_debug & PMD_DEP)
6044 		prdeps("After discard\n");
6045 #endif
6046 	while (pm_dep_head) {
6047 		rp = pm_dep_head;
6048 		if (!rp->pdr_isprop) {
6049 			ASSERT(rp->pdr_satisfied == 0);
6050 			ASSERT(pm_unresolved_deps);
6051 			pm_unresolved_deps--;
6052 		} else {
6053 			ASSERT(pm_prop_deps);
6054 			pm_prop_deps--;
6055 		}
6056 		pm_dep_head = rp->pdr_next;
6057 		if (rp->pdr_kept_count)  {
6058 			for (i = 0; i < rp->pdr_kept_count; i++) {
6059 				length = strlen(rp->pdr_kept_paths[i]) + 1;
6060 				kmem_free(rp->pdr_kept_paths[i], length);
6061 			}
6062 			kmem_free(rp->pdr_kept_paths,
6063 				rp->pdr_kept_count * sizeof (char **));
6064 		}
6065 		kmem_free(rp, rp->pdr_size);
6066 	}
6067 }
6068 
6069 
6070 static int
6071 pm_discard_dep_walk(dev_info_t *dip, void *arg)
6072 {
6073 	_NOTE(ARGUNUSED(arg))
6074 	char *pathbuf;
6075 
6076 	if (PM_GET_PM_INFO(dip) == NULL)
6077 		return (DDI_WALK_CONTINUE);
6078 	pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6079 	(void) ddi_pathname(dip, pathbuf);
6080 	pm_free_keeper(pathbuf, 0);
6081 	kmem_free(pathbuf, MAXPATHLEN);
6082 	return (DDI_WALK_CONTINUE);
6083 }
6084 
6085 static int
6086 pm_kept_walk(dev_info_t *dip, void *arg)
6087 {
6088 	_NOTE(ARGUNUSED(arg))
6089 	char *pathbuf;
6090 
6091 	pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6092 	(void) ddi_pathname(dip, pathbuf);
6093 	(void) pm_kept(pathbuf);
6094 	kmem_free(pathbuf, MAXPATHLEN);
6095 
6096 	return (DDI_WALK_CONTINUE);
6097 }
6098 
6099 static int
6100 pm_keeper_walk(dev_info_t *dip, void *arg)
6101 {
6102 	_NOTE(ARGUNUSED(arg))
6103 	char *pathbuf;
6104 
6105 	pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6106 	(void) ddi_pathname(dip, pathbuf);
6107 	(void) pm_keeper(pathbuf);
6108 	kmem_free(pathbuf, MAXPATHLEN);
6109 
6110 	return (DDI_WALK_CONTINUE);
6111 }
6112 
6113 static char *
6114 pdw_type_decode(int type)
6115 {
6116 	switch (type) {
6117 	case PM_DEP_WK_POWER_ON:
6118 		return ("power on");
6119 	case PM_DEP_WK_POWER_OFF:
6120 		return ("power off");
6121 	case PM_DEP_WK_DETACH:
6122 		return ("detach");
6123 	case PM_DEP_WK_REMOVE_DEP:
6124 		return ("remove dep");
6125 	case PM_DEP_WK_BRINGUP_SELF:
6126 		return ("bringup self");
6127 	case PM_DEP_WK_RECORD_KEEPER:
6128 		return ("add dependent");
6129 	case PM_DEP_WK_RECORD_KEEPER_PROP:
6130 		return ("add dependent property");
6131 	case PM_DEP_WK_KEPT:
6132 		return ("kept");
6133 	case PM_DEP_WK_KEEPER:
6134 		return ("keeper");
6135 	case PM_DEP_WK_ATTACH:
6136 		return ("attach");
6137 	case PM_DEP_WK_CHECK_KEPT:
6138 		return ("check kept");
6139 	case PM_DEP_WK_CPR_SUSPEND:
6140 		return ("suspend");
6141 	case PM_DEP_WK_CPR_RESUME:
6142 		return ("resume");
6143 	default:
6144 		return ("unknown");
6145 	}
6146 
6147 }
6148 
6149 static void
6150 pm_rele_dep(char *keeper)
6151 {
6152 	PMD_FUNC(pmf, "rele_dep")
6153 	pm_pdr_t *dp;
6154 	char *kept_path = NULL;
6155 	dev_info_t *kept = NULL;
6156 	int count = 0;
6157 
6158 	for (dp = pm_dep_head; dp; dp = dp->pdr_next) {
6159 		if (strcmp(dp->pdr_keeper, keeper) != 0)
6160 			continue;
6161 		for (count = 0; count < dp->pdr_kept_count; count++) {
6162 			kept_path = dp->pdr_kept_paths[count];
6163 			if (kept_path == NULL)
6164 				continue;
6165 			kept = pm_name_to_dip(kept_path, 1);
6166 			if (kept) {
6167 				PMD(PMD_KEEPS, ("%s: release kept=%s@%s(%s#%d) "
6168 				    "of keeper=%s\n", pmf, PM_DEVICE(kept),
6169 				    keeper))
6170 				ASSERT(DEVI(kept)->devi_pm_kidsupcnt > 0);
6171 				pm_rele_power(kept);
6172 				ddi_release_devi(kept);
6173 			}
6174 		}
6175 	}
6176 }
6177 
6178 /*
6179  * Called when we are just released from direct PM.  Bring ourself up
6180  * if our keeper is up since dependency is not honored while a kept
6181  * device is under direct PM.
6182  */
6183 static void
6184 pm_bring_self_up(char *keptpath)
6185 {
6186 	PMD_FUNC(pmf, "bring_self_up")
6187 	dev_info_t *kept;
6188 	dev_info_t *keeper;
6189 	pm_pdr_t *dp;
6190 	int i, j;
6191 	int up = 0, circ;
6192 
6193 	kept = pm_name_to_dip(keptpath, 1);
6194 	if (kept == NULL)
6195 		return;
6196 	PMD(PMD_KEEPS, ("%s: kept=%s@%s(%s#%d)\n", pmf, PM_DEVICE(kept)))
6197 	for (dp = pm_dep_head; dp; dp = dp->pdr_next) {
6198 		if (dp->pdr_kept_count == 0)
6199 			continue;
6200 		for (i = 0; i < dp->pdr_kept_count; i++) {
6201 			if (strcmp(dp->pdr_kept_paths[i], keptpath) != 0)
6202 				continue;
6203 			keeper = pm_name_to_dip(dp->pdr_keeper, 1);
6204 			if (keeper) {
6205 				PMD(PMD_KEEPS, ("%s: keeper=%s@%s(%s#%d)\n",
6206 				    pmf, PM_DEVICE(keeper)))
6207 				PM_LOCK_POWER(keeper, &circ);
6208 				for (j = 0; j < PM_NUMCMPTS(keeper);
6209 				    j++) {
6210 					if (PM_CURPOWER(keeper, j)) {
6211 						PMD(PMD_KEEPS, ("%s: comp="
6212 						    "%d is up\n", pmf, j))
6213 						up++;
6214 					}
6215 				}
6216 				if (up) {
6217 					if (PM_SKBU(kept))
6218 						DEVI(kept)->devi_pm_flags &=
6219 						    ~PMC_SKIP_BRINGUP;
6220 					bring_pmdep_up(kept, 1);
6221 				}
6222 				PM_UNLOCK_POWER(keeper, circ);
6223 				ddi_release_devi(keeper);
6224 			}
6225 		}
6226 	}
6227 	ddi_release_devi(kept);
6228 }
6229 
6230 static void
6231 pm_process_dep_request(pm_dep_wk_t *work)
6232 {
6233 	PMD_FUNC(pmf, "dep_req")
6234 	int ret;
6235 
6236 	PMD(PMD_DEP, ("%s: work=%s\n", pmf,
6237 	    pdw_type_decode(work->pdw_type)))
6238 	PMD(PMD_DEP, ("%s: keeper=%s, kept=%s\n", pmf,
6239 	    (work->pdw_keeper ? work->pdw_keeper : "NULL"),
6240 	    (work->pdw_kept ? work->pdw_kept : "NULL")))
6241 
6242 	switch (work->pdw_type) {
6243 	case PM_DEP_WK_POWER_ON:
6244 		/* Bring up the kept devices and put a hold on them */
6245 		bring_wekeeps_up(work->pdw_keeper);
6246 		break;
6247 	case PM_DEP_WK_POWER_OFF:
6248 		/* Release the kept devices */
6249 		pm_rele_dep(work->pdw_keeper);
6250 		break;
6251 	case PM_DEP_WK_DETACH:
6252 		pm_free_keeps(work->pdw_keeper, work->pdw_pwr);
6253 		break;
6254 	case PM_DEP_WK_REMOVE_DEP:
6255 		pm_discard_dependencies();
6256 		break;
6257 	case PM_DEP_WK_BRINGUP_SELF:
6258 		/*
6259 		 * We deferred satisfying our dependency till now, so satisfy
6260 		 * it again and bring ourselves up.
6261 		 */
6262 		pm_bring_self_up(work->pdw_kept);
6263 		break;
6264 	case PM_DEP_WK_RECORD_KEEPER:
6265 		(void) pm_record_keeper(work->pdw_kept, work->pdw_keeper, 0);
6266 		ddi_walk_devs(ddi_root_node(), pm_kept_walk, NULL);
6267 		ddi_walk_devs(ddi_root_node(), pm_keeper_walk, NULL);
6268 		break;
6269 	case PM_DEP_WK_RECORD_KEEPER_PROP:
6270 		(void) pm_record_keeper(work->pdw_kept, work->pdw_keeper, 1);
6271 		ddi_walk_devs(ddi_root_node(), pm_keeper_walk, NULL);
6272 		ddi_walk_devs(ddi_root_node(), pm_kept_walk, NULL);
6273 		break;
6274 	case PM_DEP_WK_KEPT:
6275 		ret = pm_kept(work->pdw_kept);
6276 		PMD(PMD_DEP, ("%s: PM_DEP_WK_KEPT: pm_kept returns %d\n", pmf,
6277 		    ret))
6278 		break;
6279 	case PM_DEP_WK_KEEPER:
6280 		ret = pm_keeper(work->pdw_keeper);
6281 		PMD(PMD_DEP, ("%s: PM_DEP_WK_KEEPER: pm_keeper returns %d\n",
6282 		    pmf, ret))
6283 		break;
6284 	case PM_DEP_WK_ATTACH:
6285 		ret = pm_keeper(work->pdw_keeper);
6286 		PMD(PMD_DEP, ("%s: PM_DEP_WK_ATTACH: pm_keeper returns %d\n",
6287 		    pmf, ret))
6288 		ret = pm_kept(work->pdw_kept);
6289 		PMD(PMD_DEP, ("%s: PM_DEP_WK_ATTACH: pm_kept returns %d\n",
6290 		    pmf, ret))
6291 		break;
6292 	case PM_DEP_WK_CHECK_KEPT:
6293 		ret = pm_is_kept(work->pdw_kept);
6294 		PMD(PMD_DEP, ("%s: PM_DEP_WK_CHECK_KEPT: kept=%s, ret=%d\n",
6295 		    pmf, work->pdw_kept, ret))
6296 		break;
6297 	case PM_DEP_WK_CPR_SUSPEND:
6298 		pm_discard_dependencies();
6299 		break;
6300 	case PM_DEP_WK_CPR_RESUME:
6301 		ddi_walk_devs(ddi_root_node(), pm_kept_walk, NULL);
6302 		ddi_walk_devs(ddi_root_node(), pm_keeper_walk, NULL);
6303 		break;
6304 	default:
6305 		ASSERT(0);
6306 		break;
6307 	}
6308 	/*
6309 	 * Free the work structure if the requester is not waiting
6310 	 * Otherwise it is the requester's responsiblity to free it.
6311 	 */
6312 	if (!work->pdw_wait) {
6313 		if (work->pdw_keeper)
6314 			kmem_free(work->pdw_keeper,
6315 			    strlen(work->pdw_keeper) + 1);
6316 		if (work->pdw_kept)
6317 			kmem_free(work->pdw_kept, strlen(work->pdw_kept) + 1);
6318 		kmem_free(work, sizeof (pm_dep_wk_t));
6319 	} else {
6320 		/*
6321 		 * Notify requester if it is waiting for it.
6322 		 */
6323 		work->pdw_ret = ret;
6324 		work->pdw_done = 1;
6325 		cv_signal(&work->pdw_cv);
6326 	}
6327 }
6328 
6329 /*
6330  * Process PM dependency requests.
6331  */
6332 static void
6333 pm_dep_thread(void)
6334 {
6335 	pm_dep_wk_t *work;
6336 	callb_cpr_t cprinfo;
6337 
6338 	CALLB_CPR_INIT(&cprinfo, &pm_dep_thread_lock, callb_generic_cpr,
6339 	    "pm_dep_thread");
6340 	for (;;) {
6341 		mutex_enter(&pm_dep_thread_lock);
6342 		if (pm_dep_thread_workq == NULL) {
6343 			CALLB_CPR_SAFE_BEGIN(&cprinfo);
6344 			cv_wait(&pm_dep_thread_cv, &pm_dep_thread_lock);
6345 			CALLB_CPR_SAFE_END(&cprinfo, &pm_dep_thread_lock);
6346 		}
6347 		work = pm_dep_thread_workq;
6348 		pm_dep_thread_workq = work->pdw_next;
6349 		if (pm_dep_thread_tail == work)
6350 			pm_dep_thread_tail = work->pdw_next;
6351 		mutex_exit(&pm_dep_thread_lock);
6352 		pm_process_dep_request(work);
6353 
6354 	}
6355 	/*NOTREACHED*/
6356 }
6357 
6358 /*
6359  * Set the power level of the indicated device to unknown (if it is not a
6360  * backwards compatible device), as it has just been resumed, and it won't
6361  * know if the power was removed or not. Adjust parent's kidsupcnt if necessary.
6362  */
6363 void
6364 pm_forget_power_level(dev_info_t *dip)
6365 {
6366 	dev_info_t *pdip = ddi_get_parent(dip);
6367 	int i, count = 0;
6368 
6369 	if (!PM_ISBC(dip)) {
6370 		for (i = 0; i < PM_NUMCMPTS(dip); i++)
6371 			count += (PM_CURPOWER(dip, i) == 0);
6372 
6373 		if (count && pdip && !PM_WANTS_NOTIFICATION(pdip))
6374 			e_pm_hold_rele_power(pdip, count);
6375 
6376 		/*
6377 		 * Count this as a power cycle if we care
6378 		 */
6379 		if (DEVI(dip)->devi_pm_volpmd &&
6380 		    PM_CP(dip, 0)->pmc_cur_pwr == 0)
6381 			DEVI(dip)->devi_pm_volpmd = 0;
6382 		for (i = 0; i < PM_NUMCMPTS(dip); i++)
6383 			e_pm_set_cur_pwr(dip, PM_CP(dip, i), PM_LEVEL_UNKNOWN);
6384 	}
6385 }
6386 
6387 /*
6388  * This function advises the caller whether it should make a power-off
6389  * transition at this time or not.  If the transition is not advised
6390  * at this time, the time that the next power-off transition can
6391  * be made from now is returned through "intervalp" pointer.
6392  * This function returns:
6393  *
6394  *  1  power-off advised
6395  *  0  power-off not advised, intervalp will point to seconds from
6396  *	  now that a power-off is advised.  If it is passed the number
6397  *	  of years that policy specifies the device should last,
6398  *	  a large number is returned as the time interval.
6399  *  -1  error
6400  */
6401 int
6402 pm_trans_check(struct pm_trans_data *datap, time_t *intervalp)
6403 {
6404 	PMD_FUNC(pmf, "pm_trans_check")
6405 	char dbuf[DC_SCSI_MFR_LEN];
6406 	struct pm_scsi_cycles *scp;
6407 	int service_years, service_weeks, full_years;
6408 	time_t now, service_seconds, tdiff;
6409 	time_t within_year, when_allowed;
6410 	char *ptr;
6411 	int lower_bound_cycles, upper_bound_cycles, cycles_allowed;
6412 	int cycles_diff, cycles_over;
6413 
6414 	if (datap == NULL) {
6415 		PMD(PMD_TCHECK, ("%s: NULL data pointer!\n", pmf))
6416 		return (-1);
6417 	}
6418 
6419 	if (datap->format == DC_SCSI_FORMAT) {
6420 		/*
6421 		 * Power cycles of the scsi drives are distributed
6422 		 * over 5 years with the following percentage ratio:
6423 		 *
6424 		 *	30%, 25%, 20%, 15%, and 10%
6425 		 *
6426 		 * The power cycle quota for each year is distributed
6427 		 * linearly through out the year.  The equation for
6428 		 * determining the expected cycles is:
6429 		 *
6430 		 *	e = a * (n / y)
6431 		 *
6432 		 * e = expected cycles
6433 		 * a = allocated cycles for this year
6434 		 * n = number of seconds since beginning of this year
6435 		 * y = number of seconds in a year
6436 		 *
6437 		 * Note that beginning of the year starts the day that
6438 		 * the drive has been put on service.
6439 		 *
6440 		 * If the drive has passed its expected cycles, we
6441 		 * can determine when it can start to power cycle
6442 		 * again to keep it on track to meet the 5-year
6443 		 * life expectancy.  The equation for determining
6444 		 * when to power cycle is:
6445 		 *
6446 		 *	w = y * (c / a)
6447 		 *
6448 		 * w = when it can power cycle again
6449 		 * y = number of seconds in a year
6450 		 * c = current number of cycles
6451 		 * a = allocated cycles for the year
6452 		 *
6453 		 */
6454 		char pcnt[DC_SCSI_NPY] = { 30, 55, 75, 90, 100 };
6455 
6456 		scp = &datap->un.scsi_cycles;
6457 		PMD(PMD_TCHECK, ("%s: format=%d, lifemax=%d, ncycles=%d, "
6458 		    "svc_date=%s, svc_flag=%d\n", pmf, datap->format,
6459 		    scp->lifemax, scp->ncycles, scp->svc_date, scp->flag))
6460 		if (scp->ncycles < 0 || scp->flag != 0) {
6461 			PMD(PMD_TCHECK, ("%s: ncycles < 0 || flag != 0\n", pmf))
6462 			return (-1);
6463 		}
6464 
6465 		if (scp->ncycles > scp->lifemax) {
6466 			*intervalp = (LONG_MAX / hz);
6467 			return (0);
6468 		}
6469 
6470 		/*
6471 		 * convert service date to time_t
6472 		 */
6473 		bcopy(scp->svc_date, dbuf, DC_SCSI_YEAR_LEN);
6474 		dbuf[DC_SCSI_YEAR_LEN] = '\0';
6475 		ptr = dbuf;
6476 		service_years = stoi(&ptr) - EPOCH_YEAR;
6477 		bcopy(&scp->svc_date[DC_SCSI_YEAR_LEN], dbuf,
6478 		    DC_SCSI_WEEK_LEN);
6479 		dbuf[DC_SCSI_WEEK_LEN] = '\0';
6480 
6481 		/*
6482 		 * scsi standard does not specify WW data,
6483 		 * could be (00-51) or (01-52)
6484 		 */
6485 		ptr = dbuf;
6486 		service_weeks = stoi(&ptr);
6487 		if (service_years < 0 ||
6488 		    service_weeks < 0 || service_weeks > 52) {
6489 			PMD(PMD_TCHECK, ("%s: service year %d and week %d\n",
6490 			    pmf, service_years, service_weeks))
6491 			return (-1);
6492 		}
6493 
6494 		/*
6495 		 * calculate service date in seconds-since-epoch,
6496 		 * adding one day for each leap-year.
6497 		 *
6498 		 * (years-since-epoch + 2) fixes integer truncation,
6499 		 * example: (8) leap-years during [1972, 2000]
6500 		 * (2000 - 1970) = 30;  and  (30 + 2) / 4 = 8;
6501 		 */
6502 		service_seconds = (service_years * DC_SPY) +
6503 		    (service_weeks * DC_SPW) +
6504 		    (((service_years + 2) / 4) * DC_SPD);
6505 
6506 		now = gethrestime_sec();
6507 		/*
6508 		 * since the granularity of 'svc_date' is day not second,
6509 		 * 'now' should be rounded up to full day.
6510 		 */
6511 		now = ((now + DC_SPD -1) / DC_SPD) * DC_SPD;
6512 		if (service_seconds > now) {
6513 			PMD(PMD_TCHECK, ("%s: service date (%ld) later "
6514 			    "than now (%ld)!\n", pmf, service_seconds, now))
6515 			return (-1);
6516 		}
6517 
6518 		tdiff = now - service_seconds;
6519 		PMD(PMD_TCHECK, ("%s: age is %ld sec\n", pmf, tdiff))
6520 
6521 		/*
6522 		 * NOTE - Leap years are not considered in the calculations
6523 		 * below.
6524 		 */
6525 		full_years = (tdiff / DC_SPY);
6526 		if ((full_years >= DC_SCSI_NPY) &&
6527 		    (scp->ncycles <= scp->lifemax))
6528 			return (1);
6529 
6530 		/*
6531 		 * Determine what is the normal cycle usage for the
6532 		 * device at the beginning and the end of this year.
6533 		 */
6534 		lower_bound_cycles = (!full_years) ? 0 :
6535 		    ((scp->lifemax * pcnt[full_years - 1]) / 100);
6536 		upper_bound_cycles = (scp->lifemax * pcnt[full_years]) / 100;
6537 
6538 		if (scp->ncycles <= lower_bound_cycles)
6539 			return (1);
6540 
6541 		/*
6542 		 * The linear slope that determines how many cycles
6543 		 * are allowed this year is number of seconds
6544 		 * passed this year over total number of seconds in a year.
6545 		 */
6546 		cycles_diff = (upper_bound_cycles - lower_bound_cycles);
6547 		within_year = (tdiff % DC_SPY);
6548 		cycles_allowed = lower_bound_cycles +
6549 		    (((uint64_t)cycles_diff * (uint64_t)within_year) / DC_SPY);
6550 		PMD(PMD_TCHECK, ("%s: lived %d yrs and %ld secs\n", pmf,
6551 		    full_years, within_year))
6552 		PMD(PMD_TCHECK, ("%s: # of cycles allowed %d\n", pmf,
6553 		    cycles_allowed))
6554 
6555 		if (scp->ncycles <= cycles_allowed)
6556 			return (1);
6557 
6558 		/*
6559 		 * The transition is not advised now but we can
6560 		 * determine when the next transition can be made.
6561 		 *
6562 		 * Depending on how many cycles the device has been
6563 		 * over-used, we may need to skip years with
6564 		 * different percentage quota in order to determine
6565 		 * when the next transition can be made.
6566 		 */
6567 		cycles_over = (scp->ncycles - lower_bound_cycles);
6568 		while (cycles_over > cycles_diff) {
6569 			full_years++;
6570 			if (full_years >= DC_SCSI_NPY) {
6571 				*intervalp = (LONG_MAX / hz);
6572 				return (0);
6573 			}
6574 			cycles_over -= cycles_diff;
6575 			lower_bound_cycles = upper_bound_cycles;
6576 			upper_bound_cycles =
6577 			    (scp->lifemax * pcnt[full_years]) / 100;
6578 			cycles_diff = (upper_bound_cycles - lower_bound_cycles);
6579 		}
6580 
6581 		/*
6582 		 * The linear slope that determines when the next transition
6583 		 * can be made is the relative position of used cycles within a
6584 		 * year over total number of cycles within that year.
6585 		 */
6586 		when_allowed = service_seconds + (full_years * DC_SPY) +
6587 		    (((uint64_t)DC_SPY * (uint64_t)cycles_over) / cycles_diff);
6588 		*intervalp = (when_allowed - now);
6589 		if (*intervalp > (LONG_MAX / hz))
6590 			*intervalp = (LONG_MAX / hz);
6591 		PMD(PMD_TCHECK, ("%s: no cycle is allowed in %ld secs\n", pmf,
6592 		    *intervalp))
6593 		return (0);
6594 	}
6595 
6596 	PMD(PMD_TCHECK, ("%s: unknown format!\n", pmf))
6597 	return (-1);
6598 }
6599 
6600 /*
6601  * Nexus drivers call into pm framework to indicate which child driver is about
6602  * to be installed.  In some platforms, ppm may need to configure the hardware
6603  * for successful installation of a driver.
6604  */
6605 int
6606 pm_init_child(dev_info_t *dip)
6607 {
6608 	power_req_t power_req;
6609 
6610 	ASSERT(ddi_binding_name(dip));
6611 	ASSERT(ddi_get_name_addr(dip));
6612 	pm_ppm_claim(dip);
6613 	if (pm_ppm_claimed(dip)) {	/* if ppm driver claims the node */
6614 		power_req.request_type = PMR_PPM_INIT_CHILD;
6615 		power_req.req.ppm_config_req.who = dip;
6616 		ASSERT(PPM(dip) != NULL);
6617 		return (pm_ctlops(PPM(dip), dip, DDI_CTLOPS_POWER, &power_req,
6618 		    NULL));
6619 	} else {
6620 #ifdef DEBUG
6621 		/* pass it to the default handler so we can debug things */
6622 		power_req.request_type = PMR_PPM_INIT_CHILD;
6623 		power_req.req.ppm_config_req.who = dip;
6624 		(void) pm_ctlops(NULL, dip,
6625 		    DDI_CTLOPS_POWER, &power_req, NULL);
6626 #endif
6627 	}
6628 	return (DDI_SUCCESS);
6629 }
6630 
6631 /*
6632  * Bring parent of a node that is about to be probed up to full power, and
6633  * arrange for it to stay up until pm_post_probe() or pm_post_attach() decide
6634  * it is time to let it go down again
6635  */
6636 void
6637 pm_pre_probe(dev_info_t *dip, pm_ppm_cookie_t *cp)
6638 {
6639 	int result;
6640 	power_req_t power_req;
6641 
6642 	bzero(cp, sizeof (*cp));
6643 	cp->ppc_dip = dip;
6644 
6645 	pm_ppm_claim(dip);
6646 	if (pm_ppm_claimed(dip)) {	/* if ppm driver claims the node */
6647 		power_req.request_type = PMR_PPM_PRE_PROBE;
6648 		power_req.req.ppm_config_req.who = dip;
6649 		ASSERT(PPM(dip) != NULL);
6650 		(void) pm_ctlops(PPM(dip), dip,
6651 		    DDI_CTLOPS_POWER, &power_req, &result);
6652 		cp->ppc_ppm = PPM(dip);
6653 	} else {
6654 #ifdef DEBUG
6655 		/* pass it to the default handler so we can debug things */
6656 		power_req.request_type = PMR_PPM_PRE_PROBE;
6657 		power_req.req.ppm_config_req.who = dip;
6658 		(void) pm_ctlops(NULL, dip,
6659 		    DDI_CTLOPS_POWER, &power_req, &result);
6660 #endif
6661 		cp->ppc_ppm = NULL;
6662 	}
6663 }
6664 
6665 int
6666 pm_pre_config(dev_info_t *dip, char *devnm)
6667 {
6668 	PMD_FUNC(pmf, "pre_config")
6669 	int ret;
6670 
6671 	if (MDI_VHCI(dip)) {
6672 		PMD(PMD_SET, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
6673 		ret = mdi_power(dip, MDI_PM_PRE_CONFIG, NULL, devnm, 0);
6674 		return (ret == MDI_SUCCESS ? DDI_SUCCESS : DDI_FAILURE);
6675 	} else if (!PM_GET_PM_INFO(dip))
6676 		return (DDI_SUCCESS);
6677 
6678 	PMD(PMD_SET, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
6679 	pm_hold_power(dip);
6680 	ret = pm_all_to_normal(dip, PM_CANBLOCK_BLOCK);
6681 	if (ret != DDI_SUCCESS)
6682 		pm_rele_power(dip);
6683 	return (ret);
6684 }
6685 
6686 /*
6687  * This routine is called by devfs during its walk to unconfigue a node.
6688  * If the call is due to auto mod_unloads and the dip is not at its
6689  * full power, we return DDI_FAILURE to terminate the walk, otherwise
6690  * return DDI_SUCCESS.
6691  */
6692 int
6693 pm_pre_unconfig(dev_info_t *dip, int flags, int *held, char *devnm)
6694 {
6695 	PMD_FUNC(pmf, "pre_unconfig")
6696 	int ret;
6697 
6698 	if (MDI_VHCI(dip)) {
6699 		PMD(PMD_SET, ("%s: %s@%s(%s#%d), flags=%x\n", pmf,
6700 		    PM_DEVICE(dip), flags))
6701 		ret = mdi_power(dip, MDI_PM_PRE_UNCONFIG, held, devnm, flags);
6702 		return (ret == MDI_SUCCESS ? DDI_SUCCESS : DDI_FAILURE);
6703 	} else if (!PM_GET_PM_INFO(dip))
6704 		return (DDI_SUCCESS);
6705 
6706 	PMD(PMD_SET, ("%s: %s@%s(%s#%d), flags=%x\n", pmf, PM_DEVICE(dip),
6707 	    flags))
6708 	*held = 0;
6709 
6710 	/*
6711 	 * If the dip is a leaf node, don't power it up.
6712 	 */
6713 	if (!ddi_get_child(dip))
6714 		return (DDI_SUCCESS);
6715 
6716 	/*
6717 	 * Do not power up the node if it is called due to auto-modunload.
6718 	 */
6719 	if ((flags & NDI_AUTODETACH) && !pm_all_at_normal(dip))
6720 		return (DDI_FAILURE);
6721 
6722 	pm_hold_power(dip);
6723 	*held = 1;
6724 	ret = pm_all_to_normal(dip, PM_CANBLOCK_BLOCK);
6725 	if (ret != DDI_SUCCESS) {
6726 		pm_rele_power(dip);
6727 		*held = 0;
6728 	}
6729 	return (ret);
6730 }
6731 
6732 /*
6733  * Notify ppm of attach action.  Parent is already held at full power by
6734  * probe action.
6735  */
6736 void
6737 pm_pre_attach(dev_info_t *dip, pm_ppm_cookie_t *cp, ddi_attach_cmd_t cmd)
6738 {
6739 	static char *me = "pm_pre_attach";
6740 	power_req_t power_req;
6741 	int result;
6742 
6743 	/*
6744 	 * Initialize and fill in the PPM cookie
6745 	 */
6746 	bzero(cp, sizeof (*cp));
6747 	cp->ppc_cmd = (int)cmd;
6748 	cp->ppc_ppm = PPM(dip);
6749 	cp->ppc_dip = dip;
6750 
6751 	/*
6752 	 * DDI_ATTACH and DDI_RESUME cmds need to call platform specific
6753 	 * Power Management stuff. DDI_RESUME also has to purge it's
6754 	 * powerlevel information.
6755 	 */
6756 	switch (cmd) {
6757 	case DDI_ATTACH:
6758 		if (cp->ppc_ppm) {	/* if ppm driver claims the node */
6759 			power_req.request_type = PMR_PPM_PRE_ATTACH;
6760 			power_req.req.ppm_config_req.who = dip;
6761 			ASSERT(PPM(dip));
6762 			(void) pm_ctlops(cp->ppc_ppm, dip, DDI_CTLOPS_POWER,
6763 			    &power_req, &result);
6764 		}
6765 #ifdef DEBUG
6766 		else {
6767 			power_req.request_type = PMR_PPM_PRE_ATTACH;
6768 			power_req.req.ppm_config_req.who = dip;
6769 			(void) pm_ctlops(NULL, dip,
6770 			    DDI_CTLOPS_POWER, &power_req, &result);
6771 		}
6772 #endif
6773 		break;
6774 	case DDI_RESUME:
6775 		pm_forget_power_level(dip);
6776 
6777 		if (cp->ppc_ppm) {	/* if ppm driver claims the node */
6778 			power_req.request_type = PMR_PPM_PRE_RESUME;
6779 			power_req.req.resume_req.who = cp->ppc_dip;
6780 			power_req.req.resume_req.cmd =
6781 			    (ddi_attach_cmd_t)cp->ppc_cmd;
6782 			ASSERT(PPM(cp->ppc_dip) == cp->ppc_ppm);
6783 			(void) pm_ctlops(cp->ppc_ppm, cp->ppc_dip,
6784 			    DDI_CTLOPS_POWER, &power_req, &result);
6785 		}
6786 #ifdef DEBUG
6787 		else {
6788 			power_req.request_type = PMR_PPM_PRE_RESUME;
6789 			power_req.req.resume_req.who = cp->ppc_dip;
6790 			power_req.req.resume_req.cmd =
6791 			    (ddi_attach_cmd_t)cp->ppc_cmd;
6792 			(void) pm_ctlops(NULL, cp->ppc_dip,
6793 			    DDI_CTLOPS_POWER, &power_req, &result);
6794 		}
6795 #endif
6796 		break;
6797 
6798 	case DDI_PM_RESUME:
6799 		break;
6800 
6801 	default:
6802 		panic(me);
6803 	}
6804 }
6805 
6806 /*
6807  * Nexus drivers call into pm framework to indicate which child driver is
6808  * being uninstalled.  In some platforms, ppm may need to reconfigure the
6809  * hardware since the device driver is no longer installed.
6810  */
6811 int
6812 pm_uninit_child(dev_info_t *dip)
6813 {
6814 	power_req_t power_req;
6815 
6816 	ASSERT(ddi_binding_name(dip));
6817 	ASSERT(ddi_get_name_addr(dip));
6818 	pm_ppm_claim(dip);
6819 	if (pm_ppm_claimed(dip)) {	/* if ppm driver claims the node */
6820 		power_req.request_type = PMR_PPM_UNINIT_CHILD;
6821 		power_req.req.ppm_config_req.who = dip;
6822 		ASSERT(PPM(dip));
6823 		return (pm_ctlops(PPM(dip), dip, DDI_CTLOPS_POWER, &power_req,
6824 		    NULL));
6825 	} else {
6826 #ifdef DEBUG
6827 		/* pass it to the default handler so we can debug things */
6828 		power_req.request_type = PMR_PPM_UNINIT_CHILD;
6829 		power_req.req.ppm_config_req.who = dip;
6830 		(void) pm_ctlops(NULL, dip, DDI_CTLOPS_POWER, &power_req, NULL);
6831 #endif
6832 	}
6833 	return (DDI_SUCCESS);
6834 }
6835 /*
6836  * Decrement kidsupcnt so scan can turn the parent back off if it is idle
6837  * Also notify ppm of result of probe if there is a ppm that cares
6838  */
6839 void
6840 pm_post_probe(pm_ppm_cookie_t *cp, int ret, int probe_failed)
6841 {
6842 	_NOTE(ARGUNUSED(probe_failed))
6843 	int result;
6844 	power_req_t power_req;
6845 
6846 	if (cp->ppc_ppm) {	/* if ppm driver claims the node */
6847 		power_req.request_type = PMR_PPM_POST_PROBE;
6848 		power_req.req.ppm_config_req.who = cp->ppc_dip;
6849 		power_req.req.ppm_config_req.result = ret;
6850 		ASSERT(PPM(cp->ppc_dip) == cp->ppc_ppm);
6851 		(void) pm_ctlops(cp->ppc_ppm, cp->ppc_dip, DDI_CTLOPS_POWER,
6852 		    &power_req, &result);
6853 	}
6854 #ifdef DEBUG
6855 	else {
6856 		power_req.request_type = PMR_PPM_POST_PROBE;
6857 		power_req.req.ppm_config_req.who = cp->ppc_dip;
6858 		power_req.req.ppm_config_req.result = ret;
6859 		(void) pm_ctlops(NULL, cp->ppc_dip, DDI_CTLOPS_POWER,
6860 		    &power_req, &result);
6861 	}
6862 #endif
6863 }
6864 
6865 void
6866 pm_post_config(dev_info_t *dip, char *devnm)
6867 {
6868 	PMD_FUNC(pmf, "post_config")
6869 
6870 	if (MDI_VHCI(dip)) {
6871 		PMD(PMD_SET, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
6872 		(void) mdi_power(dip, MDI_PM_POST_CONFIG, NULL, devnm, 0);
6873 		return;
6874 	} else if (!PM_GET_PM_INFO(dip))
6875 		return;
6876 
6877 	PMD(PMD_SET, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
6878 	pm_rele_power(dip);
6879 }
6880 
6881 void
6882 pm_post_unconfig(dev_info_t *dip, int held, char *devnm)
6883 {
6884 	PMD_FUNC(pmf, "post_unconfig")
6885 
6886 	if (MDI_VHCI(dip)) {
6887 		PMD(PMD_SET, ("%s: %s@%s(%s#%d), held = %d\n", pmf,
6888 		    PM_DEVICE(dip), held))
6889 		(void) mdi_power(dip, MDI_PM_POST_UNCONFIG, &held, devnm, 0);
6890 		return;
6891 	} else if (!PM_GET_PM_INFO(dip))
6892 		return;
6893 
6894 	PMD(PMD_SET, ("%s: %s@%s(%s#%d), held = %d\n", pmf, PM_DEVICE(dip),
6895 	    held))
6896 	if (!held)
6897 		return;
6898 	/*
6899 	 * We have held power in pre_unconfig, release it here.
6900 	 */
6901 	pm_rele_power(dip);
6902 }
6903 
6904 /*
6905  * Notify ppm of result of attach if there is a ppm that cares
6906  */
6907 void
6908 pm_post_attach(pm_ppm_cookie_t *cp, int ret)
6909 {
6910 	int result;
6911 	power_req_t power_req;
6912 	dev_info_t	*dip;
6913 
6914 	if (cp->ppc_cmd != DDI_ATTACH)
6915 		return;
6916 
6917 	dip = cp->ppc_dip;
6918 
6919 	if (ret == DDI_SUCCESS) {
6920 		/*
6921 		 * Attach succeeded, so proceed to doing post-attach pm tasks
6922 		 */
6923 		if (PM_GET_PM_INFO(dip) == NULL)
6924 			(void) pm_start(dip);
6925 	} else {
6926 		/*
6927 		 * Attach may have got pm started before failing
6928 		 */
6929 		pm_stop(dip);
6930 	}
6931 
6932 	if (cp->ppc_ppm) {	/* if ppm driver claims the node */
6933 		power_req.request_type = PMR_PPM_POST_ATTACH;
6934 		power_req.req.ppm_config_req.who = cp->ppc_dip;
6935 		power_req.req.ppm_config_req.result = ret;
6936 		ASSERT(PPM(cp->ppc_dip) == cp->ppc_ppm);
6937 		(void) pm_ctlops(cp->ppc_ppm, cp->ppc_dip,
6938 		    DDI_CTLOPS_POWER, &power_req, &result);
6939 	}
6940 #ifdef DEBUG
6941 	else {
6942 		power_req.request_type = PMR_PPM_POST_ATTACH;
6943 		power_req.req.ppm_config_req.who = cp->ppc_dip;
6944 		power_req.req.ppm_config_req.result = ret;
6945 		(void) pm_ctlops(NULL, cp->ppc_dip,
6946 		    DDI_CTLOPS_POWER, &power_req, &result);
6947 	}
6948 #endif
6949 }
6950 
6951 /*
6952  * Notify ppm of attach action.  Parent is already held at full power by
6953  * probe action.
6954  */
6955 void
6956 pm_pre_detach(dev_info_t *dip, ddi_detach_cmd_t cmd, pm_ppm_cookie_t *cp)
6957 {
6958 	int result;
6959 	power_req_t power_req;
6960 
6961 	bzero(cp, sizeof (*cp));
6962 	cp->ppc_dip = dip;
6963 	cp->ppc_cmd = (int)cmd;
6964 
6965 	switch (cmd) {
6966 	case DDI_DETACH:
6967 		pm_detaching(dip);		/* suspend pm while detaching */
6968 		if (pm_ppm_claimed(dip)) {	/* if ppm driver claims node */
6969 			power_req.request_type = PMR_PPM_PRE_DETACH;
6970 			power_req.req.ppm_config_req.who = dip;
6971 			ASSERT(PPM(dip));
6972 			(void) pm_ctlops(PPM(dip), dip, DDI_CTLOPS_POWER,
6973 			    &power_req, &result);
6974 			cp->ppc_ppm = PPM(dip);
6975 		} else {
6976 #ifdef DEBUG
6977 			/* pass to the default handler so we can debug things */
6978 			power_req.request_type = PMR_PPM_PRE_DETACH;
6979 			power_req.req.ppm_config_req.who = dip;
6980 			(void) pm_ctlops(NULL, dip,
6981 			    DDI_CTLOPS_POWER, &power_req, &result);
6982 #endif
6983 			cp->ppc_ppm = NULL;
6984 		}
6985 		break;
6986 
6987 	default:
6988 		break;
6989 	}
6990 }
6991 
6992 /*
6993  * Dip is either a leaf node that exported "no-involuntary-power-cycles" prop.,
6994  * (if devi_pm_noinvol count is 0) or an ancestor of such a node.  We need to
6995  * make an entry to record the details, which includes certain flag settings.
6996  */
6997 static void
6998 pm_record_invol_path(char *path, int flags, int noinvolpm, int volpmd,
6999     int wasvolpmd, major_t major)
7000 {
7001 	PMD_FUNC(pmf, "record_invol_path")
7002 	major_t pm_path_to_major(char *);
7003 	size_t plen;
7004 	pm_noinvol_t *ip, *np, *pp;
7005 	pp = NULL;
7006 
7007 	plen = strlen(path) + 1;
7008 	np = kmem_zalloc(sizeof (*np), KM_SLEEP);
7009 	np->ni_size = plen;
7010 	np->ni_path = kmem_alloc(plen, KM_SLEEP);
7011 	np->ni_noinvolpm = noinvolpm;
7012 	np->ni_volpmd = volpmd;
7013 	np->ni_wasvolpmd = wasvolpmd;
7014 	np->ni_flags = flags;
7015 	(void) strcpy(np->ni_path, path);
7016 	/*
7017 	 * If we haven't actually seen the node attached, it is hard to figure
7018 	 * out its major.  If we could hold the node by path, we would be much
7019 	 * happier here.
7020 	 */
7021 	if (major == (major_t)-1) {
7022 		np->ni_major = pm_path_to_major(path);
7023 	} else {
7024 		np->ni_major = major;
7025 	}
7026 	rw_enter(&pm_noinvol_rwlock, RW_WRITER);
7027 	for (ip = pm_noinvol_head; ip; pp = ip, ip = ip->ni_next) {
7028 		int comp = strcmp(path, ip->ni_path);
7029 		if (comp < 0) {
7030 			PMD(PMD_NOINVOL, ("%s: %s insert before %s\n",
7031 			    pmf, path, ip->ni_path))
7032 			/* insert before current entry */
7033 			np->ni_next = ip;
7034 			if (pp) {
7035 				pp->ni_next = np;
7036 			} else {
7037 				pm_noinvol_head = np;
7038 			}
7039 			rw_exit(&pm_noinvol_rwlock);
7040 #ifdef DEBUG
7041 			if (pm_debug & PMD_NOINVOL)
7042 				pr_noinvol("record_invol_path exit0");
7043 #endif
7044 			return;
7045 		} else if (comp == 0) {
7046 			panic("%s already in pm_noinvol list", path);
7047 		}
7048 	}
7049 	/*
7050 	 * If we did not find an entry in the list that this should go before,
7051 	 * then it must go at the end
7052 	 */
7053 	if (pp) {
7054 		PMD(PMD_NOINVOL, ("%s: %s append after %s\n", pmf, path,
7055 		    pp->ni_path))
7056 		ASSERT(pp->ni_next == 0);
7057 		pp->ni_next = np;
7058 	} else {
7059 		PMD(PMD_NOINVOL, ("%s: %s added to end-of-list\n", pmf, path))
7060 		ASSERT(!pm_noinvol_head);
7061 		pm_noinvol_head = np;
7062 	}
7063 	rw_exit(&pm_noinvol_rwlock);
7064 #ifdef DEBUG
7065 	if (pm_debug & PMD_NOINVOL)
7066 		pr_noinvol("record_invol_path exit");
7067 #endif
7068 }
7069 
7070 void
7071 pm_record_invol(dev_info_t *dip)
7072 {
7073 	char *pathbuf;
7074 	int pm_all_components_off(dev_info_t *);
7075 	int volpmd = (PM_NUMCMPTS(dip) > 0) && pm_all_components_off(dip);
7076 
7077 	pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
7078 	(void) ddi_pathname(dip, pathbuf);
7079 
7080 	pm_record_invol_path(pathbuf, (DEVI(dip)->devi_pm_flags &
7081 	    (PMC_NO_INVOL | PMC_CONSOLE_FB)), DEVI(dip)->devi_pm_noinvolpm,
7082 	    DEVI(dip)->devi_pm_volpmd, volpmd, PM_MAJOR(dip));
7083 
7084 	/*
7085 	 * If this child's detach will be holding up its ancestors, then we
7086 	 * allow for an exception to that if all children of this type have
7087 	 * gone down voluntarily.
7088 	 * Now walk down the tree incrementing devi_pm_noinvolpm
7089 	 */
7090 	(void) pm_noinvol_update(PM_BP_NOINVOL_DETACH, 0, volpmd, pathbuf,
7091 	    dip);
7092 	kmem_free(pathbuf, MAXPATHLEN);
7093 }
7094 
7095 void
7096 pm_post_detach(pm_ppm_cookie_t *cp, int ret)
7097 {
7098 	dev_info_t *dip = cp->ppc_dip;
7099 	int result;
7100 	power_req_t power_req;
7101 
7102 	switch (cp->ppc_cmd) {
7103 	case DDI_DETACH:
7104 		if (cp->ppc_ppm) {	/* if ppm driver claims the node */
7105 			power_req.request_type = PMR_PPM_POST_DETACH;
7106 			power_req.req.ppm_config_req.who = cp->ppc_dip;
7107 			power_req.req.ppm_config_req.result = ret;
7108 			ASSERT(PPM(cp->ppc_dip) == cp->ppc_ppm);
7109 			(void) pm_ctlops(cp->ppc_ppm, cp->ppc_dip,
7110 			    DDI_CTLOPS_POWER, &power_req, &result);
7111 		}
7112 #ifdef DEBUG
7113 		else {
7114 			power_req.request_type = PMR_PPM_POST_DETACH;
7115 			power_req.req.ppm_config_req.who = cp->ppc_dip;
7116 			power_req.req.ppm_config_req.result = ret;
7117 			(void) pm_ctlops(NULL, cp->ppc_dip,
7118 			    DDI_CTLOPS_POWER, &power_req, &result);
7119 		}
7120 #endif
7121 		if (ret == DDI_SUCCESS) {
7122 			/*
7123 			 * For hotplug detach we assume it is *really* gone
7124 			 */
7125 			if (cp->ppc_cmd == DDI_DETACH &&
7126 			    ((DEVI(dip)->devi_pm_flags &
7127 			    (PMC_NO_INVOL | PMC_CONSOLE_FB)) ||
7128 			    DEVI(dip)->devi_pm_noinvolpm))
7129 				pm_record_invol(dip);
7130 			DEVI(dip)->devi_pm_flags &=
7131 			    ~(PMC_NO_INVOL | PMC_NOINVOL_DONE);
7132 
7133 			/*
7134 			 * If console fb is detaching, then we don't need to
7135 			 * worry any more about it going off (pm_detaching has
7136 			 * brought up all components)
7137 			 */
7138 			if (PM_IS_CFB(dip)) {
7139 				mutex_enter(&pm_cfb_lock);
7140 				ASSERT(cfb_dip_detaching);
7141 				ASSERT(cfb_dip == NULL);
7142 				ASSERT(pm_cfb_comps_off == 0);
7143 				cfb_dip_detaching = NULL;
7144 				mutex_exit(&pm_cfb_lock);
7145 			}
7146 			pm_stop(dip);	/* make it permanent */
7147 		} else {
7148 			if (PM_IS_CFB(dip)) {
7149 				mutex_enter(&pm_cfb_lock);
7150 				ASSERT(cfb_dip_detaching);
7151 				ASSERT(cfb_dip == NULL);
7152 				ASSERT(pm_cfb_comps_off == 0);
7153 				cfb_dip = cfb_dip_detaching;
7154 				cfb_dip_detaching = NULL;
7155 				mutex_exit(&pm_cfb_lock);
7156 			}
7157 			pm_detach_failed(dip);	/* resume power management */
7158 		}
7159 		break;
7160 	case DDI_PM_SUSPEND:
7161 		break;
7162 	case DDI_SUSPEND:
7163 		break;				/* legal, but nothing to do */
7164 	default:
7165 #ifdef DEBUG
7166 		panic("pm_post_detach: unrecognized cmd %d for detach",
7167 		    cp->ppc_cmd);
7168 		/*NOTREACHED*/
7169 #else
7170 		break;
7171 #endif
7172 	}
7173 }
7174 
7175 /*
7176  * Called after vfs_mountroot has got the clock started to fix up timestamps
7177  * that were set when root bush drivers attached.  hresttime was 0 then, so the
7178  * devices look busy but have a 0 busycnt
7179  */
7180 int
7181 pm_adjust_timestamps(dev_info_t *dip, void *arg)
7182 {
7183 	_NOTE(ARGUNUSED(arg))
7184 
7185 	pm_info_t *info = PM_GET_PM_INFO(dip);
7186 	struct pm_component *cp;
7187 	int i;
7188 
7189 	if (!info)
7190 		return (DDI_WALK_CONTINUE);
7191 	PM_LOCK_BUSY(dip);
7192 	for (i = 0; i < PM_NUMCMPTS(dip); i++) {
7193 		cp = PM_CP(dip, i);
7194 		if (cp->pmc_timestamp == 0 && cp->pmc_busycount == 0)
7195 			cp->pmc_timestamp = gethrestime_sec();
7196 	}
7197 	PM_UNLOCK_BUSY(dip);
7198 	return (DDI_WALK_CONTINUE);
7199 }
7200 
7201 /*
7202  * Called at attach time to see if the device being attached has a record in
7203  * the no involuntary power cycles list.  If so, we do some bookkeeping on the
7204  * parents and set a flag in the dip
7205  */
7206 void
7207 pm_noinvol_specd(dev_info_t *dip)
7208 {
7209 	PMD_FUNC(pmf, "noinvol_specd")
7210 	char *pathbuf;
7211 	pm_noinvol_t *ip, *pp = NULL;
7212 	int wasvolpmd;
7213 	int found = 0;
7214 
7215 	if (DEVI(dip)->devi_pm_flags & PMC_NOINVOL_DONE)
7216 		return;
7217 	DEVI(dip)->devi_pm_flags |=  PMC_NOINVOL_DONE;
7218 	pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
7219 	(void) ddi_pathname(dip, pathbuf);
7220 
7221 	PM_LOCK_DIP(dip);
7222 	DEVI(dip)->devi_pm_volpmd = 0;
7223 	DEVI(dip)->devi_pm_noinvolpm = 0;
7224 	rw_enter(&pm_noinvol_rwlock, RW_READER);
7225 	for (ip = pm_noinvol_head; ip; pp = ip, ip = ip->ni_next) {
7226 		PMD(PMD_NOINVOL, ("%s: comparing '%s' to '%s'\n",
7227 		    pmf, pathbuf, ip->ni_path))
7228 		if (strcmp(pathbuf, ip->ni_path) == 0) {
7229 			found++;
7230 			break;
7231 		}
7232 	}
7233 	rw_exit(&pm_noinvol_rwlock);
7234 	if (!found) {
7235 		PM_UNLOCK_DIP(dip);
7236 		kmem_free(pathbuf, MAXPATHLEN);
7237 		return;
7238 	}
7239 	rw_enter(&pm_noinvol_rwlock, RW_WRITER);
7240 	pp = NULL;
7241 	for (ip = pm_noinvol_head; ip; pp = ip, ip = ip->ni_next) {
7242 		PMD(PMD_NOINVOL, ("%s: comparing '%s' to '%s'\n",
7243 		    pmf, pathbuf, ip->ni_path))
7244 		if (strcmp(pathbuf, ip->ni_path) == 0) {
7245 			ip->ni_flags &= ~PMC_DRIVER_REMOVED;
7246 			DEVI(dip)->devi_pm_flags |= ip->ni_flags;
7247 			/*
7248 			 * Handle special case of console fb
7249 			 */
7250 			if (PM_IS_CFB(dip)) {
7251 				mutex_enter(&pm_cfb_lock);
7252 				cfb_dip = dip;
7253 				PMD(PMD_CFB, ("%s: %s@%s(%s#%d) setting "
7254 				    "cfb_dip\n", pmf, PM_DEVICE(dip)))
7255 				mutex_exit(&pm_cfb_lock);
7256 			}
7257 			DEVI(dip)->devi_pm_noinvolpm = ip->ni_noinvolpm;
7258 			ASSERT((DEVI(dip)->devi_pm_flags &
7259 			    (PMC_NO_INVOL | PMC_CONSOLE_FB)) ||
7260 			    DEVI(dip)->devi_pm_noinvolpm);
7261 			DEVI(dip)->devi_pm_volpmd = ip->ni_volpmd;
7262 			PMD(PMD_NOINVOL, ("%s: noinvol=%d, volpmd=%d, "
7263 			    "wasvolpmd=%d, flags=%x, path=%s\n", pmf,
7264 			    ip->ni_noinvolpm, ip->ni_volpmd,
7265 			    ip->ni_wasvolpmd, ip->ni_flags, ip->ni_path))
7266 			/*
7267 			 * free the entry in hopes the list will now be empty
7268 			 * and we won't have to search it any more until the
7269 			 * device detaches
7270 			 */
7271 			if (pp) {
7272 				PMD(PMD_NOINVOL, ("%s: free %s, prev %s\n",
7273 				    pmf, ip->ni_path, pp->ni_path))
7274 				pp->ni_next = ip->ni_next;
7275 			} else {
7276 				PMD(PMD_NOINVOL, ("%s: free %s head\n",
7277 				    pmf, ip->ni_path))
7278 				ASSERT(pm_noinvol_head == ip);
7279 				pm_noinvol_head = ip->ni_next;
7280 			}
7281 			PM_UNLOCK_DIP(dip);
7282 			wasvolpmd = ip->ni_wasvolpmd;
7283 			rw_exit(&pm_noinvol_rwlock);
7284 			kmem_free(ip->ni_path, ip->ni_size);
7285 			kmem_free(ip, sizeof (*ip));
7286 			/*
7287 			 * Now walk up the tree decrementing devi_pm_noinvolpm
7288 			 * (and volpmd if appropriate)
7289 			 */
7290 			(void) pm_noinvol_update(PM_BP_NOINVOL_ATTACH, 0,
7291 			    wasvolpmd, pathbuf, dip);
7292 #ifdef DEBUG
7293 			if (pm_debug & PMD_NOINVOL)
7294 				pr_noinvol("noinvol_specd exit");
7295 #endif
7296 			kmem_free(pathbuf, MAXPATHLEN);
7297 			return;
7298 		}
7299 	}
7300 	kmem_free(pathbuf, MAXPATHLEN);
7301 	rw_exit(&pm_noinvol_rwlock);
7302 	PM_UNLOCK_DIP(dip);
7303 }
7304 
7305 int
7306 pm_all_components_off(dev_info_t *dip)
7307 {
7308 	int i;
7309 	pm_component_t *cp;
7310 
7311 	for (i = 0; i < PM_NUMCMPTS(dip); i++) {
7312 		cp = PM_CP(dip, i);
7313 		if (cp->pmc_cur_pwr == PM_LEVEL_UNKNOWN ||
7314 		    cp->pmc_comp.pmc_lvals[cp->pmc_cur_pwr])
7315 			return (0);
7316 	}
7317 	return (1);	/* all off */
7318 }
7319 
7320 /*
7321  * Make sure that all "no involuntary power cycles" devices are attached.
7322  * Called before doing a cpr suspend to make sure the driver has a say about
7323  * the power cycle
7324  */
7325 int
7326 pm_reattach_noinvol(void)
7327 {
7328 	PMD_FUNC(pmf, "reattach_noinvol")
7329 	pm_noinvol_t *ip;
7330 	char *path;
7331 	dev_info_t *dip;
7332 
7333 	/*
7334 	 * Prevent the modunload thread from unloading any modules until we
7335 	 * have completely stopped all kernel threads.
7336 	 */
7337 	modunload_disable();
7338 	for (ip = pm_noinvol_head; ip; ip = ip->ni_next) {
7339 		/*
7340 		 * Forget we'v ever seen any entry
7341 		 */
7342 		ip->ni_persistent = 0;
7343 	}
7344 restart:
7345 	rw_enter(&pm_noinvol_rwlock, RW_READER);
7346 	for (ip = pm_noinvol_head; ip; ip = ip->ni_next) {
7347 		major_t maj;
7348 		maj = ip->ni_major;
7349 		path = ip->ni_path;
7350 		if (path != NULL && !(ip->ni_flags & PMC_DRIVER_REMOVED)) {
7351 			if (ip->ni_persistent) {
7352 				/*
7353 				 * If we weren't able to make this entry
7354 				 * go away, then we give up, as
7355 				 * holding/attaching the driver ought to have
7356 				 * resulted in this entry being deleted
7357 				 */
7358 				PMD(PMD_NOINVOL, ("%s: can't reattach %s "
7359 				    "(%s|%d)\n", pmf, ip->ni_path,
7360 				    ddi_major_to_name(maj), (int)maj))
7361 				cmn_err(CE_WARN, "cpr: unable to reattach %s ",
7362 				    ip->ni_path);
7363 				modunload_enable();
7364 				rw_exit(&pm_noinvol_rwlock);
7365 				return (0);
7366 			}
7367 			ip->ni_persistent++;
7368 			rw_exit(&pm_noinvol_rwlock);
7369 			PMD(PMD_NOINVOL, ("%s: holding %s\n", pmf, path))
7370 			dip = e_ddi_hold_devi_by_path(path, 0);
7371 			if (dip == NULL) {
7372 				PMD(PMD_NOINVOL, ("%s: can't hold (%s|%d)\n",
7373 				    pmf, path, (int)maj))
7374 				cmn_err(CE_WARN, "cpr: unable to hold %s "
7375 				    "driver", path);
7376 				modunload_enable();
7377 				return (0);
7378 			} else {
7379 				PMD(PMD_DHR, ("%s: release %s\n", pmf, path))
7380 				/*
7381 				 * Since the modunload thread is stopped, we
7382 				 * don't have to keep the driver held, which
7383 				 * saves a ton of bookkeeping
7384 				 */
7385 				ddi_release_devi(dip);
7386 				goto restart;
7387 			}
7388 		} else {
7389 			PMD(PMD_NOINVOL, ("%s: skip %s; unknown major\n",
7390 			    pmf, ip->ni_path))
7391 			continue;
7392 		}
7393 	}
7394 	rw_exit(&pm_noinvol_rwlock);
7395 	return (1);
7396 }
7397 
7398 void
7399 pm_reattach_noinvol_fini(void)
7400 {
7401 	modunload_enable();
7402 }
7403 
7404 /*
7405  * Display pm support code
7406  */
7407 
7408 
7409 /*
7410  * console frame-buffer power-mgmt gets enabled when debugging
7411  * services are not present or console fbpm override is set
7412  */
7413 void
7414 pm_cfb_setup(const char *stdout_path)
7415 {
7416 	PMD_FUNC(pmf, "cfb_setup")
7417 	extern int obpdebug;
7418 	char *devname;
7419 	dev_info_t *dip;
7420 	int devname_len;
7421 	extern dev_info_t *fbdip;
7422 
7423 	/*
7424 	 * By virtue of this function being called (from consconfig),
7425 	 * we know stdout is a framebuffer.
7426 	 */
7427 	stdout_is_framebuffer = 1;
7428 
7429 	if (obpdebug || (boothowto & RB_DEBUG)) {
7430 		if (pm_cfb_override == 0) {
7431 			/*
7432 			 * Console is frame buffer, but we want to suppress
7433 			 * pm on it because of debugging setup
7434 			 */
7435 			pm_cfb_enabled = 0;
7436 			cmn_err(CE_NOTE, "Kernel debugger present: disabling "
7437 			    "console power management.");
7438 			/*
7439 			 * however, we still need to know which is the console
7440 			 * fb in order to suppress pm on it
7441 			 */
7442 		} else {
7443 			cmn_err(CE_WARN, "Kernel debugger present: see "
7444 			    "kmdb(1M) for interaction with power management.");
7445 		}
7446 	}
7447 #ifdef DEBUG
7448 	/*
7449 	 * IF console is fb and is power managed, don't do prom_printfs from
7450 	 * pm debug macro
7451 	 */
7452 	if (pm_cfb_enabled) {
7453 		if (pm_debug)
7454 			prom_printf("pm debug output will be to log only\n");
7455 		pm_divertdebug++;
7456 	}
7457 #endif
7458 	devname = i_ddi_strdup((char *)stdout_path, KM_SLEEP);
7459 	devname_len = strlen(devname) + 1;
7460 	PMD(PMD_CFB, ("%s: stripped %s\n", pmf, devname))
7461 	/* if the driver is attached */
7462 	if ((dip = fbdip) != NULL) {
7463 		PMD(PMD_CFB, ("%s: attached: %s@%s(%s#%d)\n", pmf,
7464 		    PM_DEVICE(dip)))
7465 		/*
7466 		 * We set up here as if the driver were power manageable in case
7467 		 * we get a later attach of a pm'able driver (which would result
7468 		 * in a panic later)
7469 		 */
7470 		cfb_dip = dip;
7471 		DEVI(dip)->devi_pm_flags |= (PMC_CONSOLE_FB | PMC_NO_INVOL);
7472 		PMD(PMD_CFB, ("%s: cfb_dip -> %s@%s(%s#%d)\n", pmf,
7473 		    PM_DEVICE(dip)))
7474 #ifdef DEBUG
7475 		if (!(PM_GET_PM_INFO(dip) != NULL && PM_NUMCMPTS(dip))) {
7476 			PMD(PMD_CFB, ("%s: %s@%s(%s#%d) not power-managed\n",
7477 			    pmf, PM_DEVICE(dip)))
7478 		}
7479 #endif
7480 	} else {
7481 		char *ep;
7482 		PMD(PMD_CFB, ("%s: pntd %s failed\n", pmf, devname))
7483 		pm_record_invol_path(devname,
7484 		    (PMC_CONSOLE_FB | PMC_NO_INVOL), 1, 0, 0,
7485 		    (major_t)-1);
7486 		for (ep = strrchr(devname, '/'); ep != devname;
7487 		    ep = strrchr(devname, '/')) {
7488 			PMD(PMD_CFB, ("%s: devname %s\n", pmf, devname))
7489 			*ep = '\0';
7490 			dip = pm_name_to_dip(devname, 0);
7491 			if (dip != NULL) {
7492 				/*
7493 				 * Walk up the tree incrementing
7494 				 * devi_pm_noinvolpm
7495 				 */
7496 				(void) pm_noinvol_update(PM_BP_NOINVOL_CFB,
7497 				    0, 0, devname, dip);
7498 				break;
7499 			} else {
7500 				pm_record_invol_path(devname,
7501 				    PMC_NO_INVOL, 1, 0, 0, (major_t)-1);
7502 			}
7503 		}
7504 	}
7505 	kmem_free(devname, devname_len);
7506 }
7507 
7508 void
7509 pm_cfb_rele(void)
7510 {
7511 	mutex_enter(&pm_cfb_lock);
7512 	/*
7513 	 * this call isn't using the console any  more, it is ok to take it
7514 	 * down if the count goes to 0
7515 	 */
7516 	cfb_inuse--;
7517 	mutex_exit(&pm_cfb_lock);
7518 }
7519 
7520 /*
7521  * software interrupt handler for fbpm; this function exists because we can't
7522  * bring up the frame buffer power from above lock level.  So if we need to,
7523  * we instead schedule a softint that runs this routine and takes us into
7524  * debug_enter (a bit delayed from the original request, but avoiding a panic).
7525  */
7526 static uint_t
7527 pm_cfb_softint(caddr_t int_handler_arg)
7528 {
7529 	_NOTE(ARGUNUSED(int_handler_arg))
7530 	int rval = DDI_INTR_UNCLAIMED;
7531 
7532 	mutex_enter(&pm_cfb_lock);
7533 	if (pm_soft_pending) {
7534 		mutex_exit(&pm_cfb_lock);
7535 		debug_enter((char *)NULL);
7536 		/* acquired in debug_enter before calling pm_cfb_trigger */
7537 		pm_cfb_rele();
7538 		mutex_enter(&pm_cfb_lock);
7539 		pm_soft_pending = 0;
7540 		mutex_exit(&pm_cfb_lock);
7541 		rval = DDI_INTR_CLAIMED;
7542 	} else
7543 		mutex_exit(&pm_cfb_lock);
7544 
7545 	return (rval);
7546 }
7547 
7548 void
7549 pm_cfb_setup_intr(void)
7550 {
7551 	PMD_FUNC(pmf, "cfb_setup_intr")
7552 	extern void prom_set_outfuncs(void (*)(void), void (*)(void));
7553 	void pm_cfb_check_and_powerup(void);
7554 
7555 	if (!stdout_is_framebuffer) {
7556 		PMD(PMD_CFB, ("%s: console not fb\n", pmf))
7557 		return;
7558 	}
7559 	mutex_init(&pm_cfb_lock, NULL, MUTEX_SPIN, (void *)ipltospl(SPL8));
7560 #ifdef DEBUG
7561 	mutex_init(&pm_debug_lock, NULL, MUTEX_SPIN, (void *)ipltospl(SPL8));
7562 #endif
7563 	/*
7564 	 * setup software interrupt handler
7565 	 */
7566 	if (ddi_add_softintr(ddi_root_node(), DDI_SOFTINT_HIGH, &pm_soft_id,
7567 	    NULL, NULL, pm_cfb_softint, NULL) != DDI_SUCCESS)
7568 		panic("pm: unable to register soft intr.");
7569 
7570 	prom_set_outfuncs(pm_cfb_check_and_powerup, pm_cfb_rele);
7571 }
7572 
7573 /*
7574  * Checks to see if it is safe to write to the console wrt power management
7575  * (i.e. if the console is a framebuffer, then it must be at full power)
7576  * returns 1 when power is off (power-up is needed)
7577  * returns 0 when power is on (power-up not needed)
7578  */
7579 int
7580 pm_cfb_check_and_hold(void)
7581 {
7582 	/*
7583 	 * cfb_dip is set iff console is a power manageable frame buffer
7584 	 * device
7585 	 */
7586 	extern int modrootloaded;
7587 
7588 	mutex_enter(&pm_cfb_lock);
7589 	cfb_inuse++;
7590 	ASSERT(cfb_inuse);	/* wrap? */
7591 	if (modrootloaded && cfb_dip) {
7592 		/*
7593 		 * don't power down the frame buffer, the prom is using it
7594 		 */
7595 		if (pm_cfb_comps_off) {
7596 			mutex_exit(&pm_cfb_lock);
7597 			return (1);
7598 		}
7599 	}
7600 	mutex_exit(&pm_cfb_lock);
7601 	return (0);
7602 }
7603 
7604 /*
7605  * turn on cfb power (which is known to be off).
7606  * Must be called below lock level!
7607  */
7608 void
7609 pm_cfb_powerup(void)
7610 {
7611 	pm_info_t *info;
7612 	int norm;
7613 	int ccount, ci;
7614 	int unused;
7615 #ifdef DEBUG
7616 	/*
7617 	 * Can't reenter prom_prekern, so suppress pm debug messages
7618 	 * (still go to circular buffer).
7619 	 */
7620 	mutex_enter(&pm_debug_lock);
7621 	pm_divertdebug++;
7622 	mutex_exit(&pm_debug_lock);
7623 #endif
7624 	info = PM_GET_PM_INFO(cfb_dip);
7625 	ASSERT(info);
7626 
7627 	ccount = PM_NUMCMPTS(cfb_dip);
7628 	for (ci = 0; ci < ccount; ci++) {
7629 		norm = pm_get_normal_power(cfb_dip, ci);
7630 		(void) pm_set_power(cfb_dip, ci, norm, PM_LEVEL_UPONLY,
7631 		    PM_CANBLOCK_BYPASS, 0, &unused);
7632 	}
7633 #ifdef DEBUG
7634 	mutex_enter(&pm_debug_lock);
7635 	pm_divertdebug--;
7636 	mutex_exit(&pm_debug_lock);
7637 #endif
7638 }
7639 
7640 /*
7641  * Check if the console framebuffer is powered up.  If not power it up.
7642  * Note: Calling pm_cfb_check_and_hold has put a hold on the power state which
7643  * must be released by calling pm_cfb_rele when the console fb operation
7644  * is completed.
7645  */
7646 void
7647 pm_cfb_check_and_powerup(void)
7648 {
7649 	if (pm_cfb_check_and_hold())
7650 		pm_cfb_powerup();
7651 }
7652 
7653 /*
7654  * Trigger a low level interrupt to power up console frame buffer.
7655  */
7656 void
7657 pm_cfb_trigger(void)
7658 {
7659 	if (cfb_dip == NULL)
7660 		return;
7661 
7662 	mutex_enter(&pm_cfb_lock);
7663 	/*
7664 	 * If machine appears to be hung, pulling the keyboard connector of
7665 	 * the console will cause a high level interrupt and go to debug_enter.
7666 	 * But, if the fb is powered down, this routine will be called to bring
7667 	 * it up (by generating a softint to do the work).  If soft interrupts
7668 	 * are not running, and the keyboard connector is pulled again, the
7669 	 * following code detects this condition and calls panic which allows
7670 	 * the fb to be brought up from high level.
7671 	 *
7672 	 * If two nearly simultaneous calls to debug_enter occur (both from
7673 	 * high level) the code described above will cause a panic.
7674 	 */
7675 	if (lbolt <= pm_soft_pending) {
7676 		panicstr = "pm_cfb_trigger: lbolt not advancing";
7677 		panic(panicstr);	/* does a power up at any intr level */
7678 		/* NOTREACHED */
7679 	}
7680 	pm_soft_pending = lbolt;
7681 	mutex_exit(&pm_cfb_lock);
7682 	ddi_trigger_softintr(pm_soft_id);
7683 }
7684 
7685 major_t
7686 pm_path_to_major(char *path)
7687 {
7688 	PMD_FUNC(pmf, "path_to_major")
7689 	char *np, *ap, *bp;
7690 	major_t ret;
7691 	size_t len;
7692 	static major_t i_path_to_major(char *, char *);
7693 
7694 	PMD(PMD_NOINVOL, ("%s: %s\n", pmf, path))
7695 
7696 	np = strrchr(path, '/');
7697 	if (np != NULL)
7698 		np++;
7699 	else
7700 		np = path;
7701 	len = strlen(np) + 1;
7702 	bp = kmem_alloc(len, KM_SLEEP);
7703 	(void) strcpy(bp, np);
7704 	if ((ap = strchr(bp, '@')) != NULL) {
7705 		*ap = '\0';
7706 	}
7707 	PMD(PMD_NOINVOL, ("%s: %d\n", pmf, ddi_name_to_major(np)))
7708 	ret = i_path_to_major(path, np);
7709 	kmem_free(bp, len);
7710 	return (ret);
7711 }
7712 
7713 #ifdef DEBUG
7714 
7715 char *pm_msgp;
7716 char *pm_bufend;
7717 char *pm_msgbuf = NULL;
7718 int   pm_logpages = 2;
7719 
7720 #define	PMLOGPGS	pm_logpages
7721 
7722 /*PRINTFLIKE1*/
7723 void
7724 pm_log(const char *fmt, ...)
7725 {
7726 	va_list adx;
7727 	size_t size;
7728 
7729 	mutex_enter(&pm_debug_lock);
7730 	if (pm_msgbuf == NULL) {
7731 		pm_msgbuf = kmem_zalloc(mmu_ptob(PMLOGPGS), KM_SLEEP);
7732 		pm_bufend = pm_msgbuf + mmu_ptob(PMLOGPGS) - 1;
7733 		pm_msgp = pm_msgbuf;
7734 	}
7735 	va_start(adx, fmt);
7736 	size = vsnprintf(NULL, 0, fmt, adx) + 1;
7737 	va_end(adx);
7738 	va_start(adx, fmt);
7739 	if (size > (pm_bufend - pm_msgp)) {		/* wraps */
7740 		bzero(pm_msgp, pm_bufend - pm_msgp);
7741 		(void) vsnprintf(pm_msgbuf, size, fmt, adx);
7742 		if (!pm_divertdebug)
7743 			prom_printf("%s", pm_msgp);
7744 		pm_msgp = pm_msgbuf + size;
7745 	} else {
7746 		(void) vsnprintf(pm_msgp, size, fmt, adx);
7747 		if (!pm_divertdebug)
7748 			prom_printf("%s", pm_msgp);
7749 		pm_msgp += size;
7750 	}
7751 	va_end(adx);
7752 	mutex_exit(&pm_debug_lock);
7753 }
7754 #endif	/* DEBUG */
7755 
7756 /*
7757  * We want to save the state of any directly pm'd devices over the suspend/
7758  * resume process so that we can put them back the way the controlling
7759  * process left them.
7760  */
7761 void
7762 pm_save_direct_levels(void)
7763 {
7764 	pm_processes_stopped = 1;
7765 	ddi_walk_devs(ddi_root_node(), pm_save_direct_lvl_walk, 0);
7766 }
7767 
7768 static int
7769 pm_save_direct_lvl_walk(dev_info_t *dip, void *arg)
7770 {
7771 	_NOTE(ARGUNUSED(arg))
7772 	int i;
7773 	int *ip;
7774 	pm_info_t *info = PM_GET_PM_INFO(dip);
7775 
7776 	if (!info)
7777 		return (DDI_WALK_CONTINUE);
7778 
7779 	if (PM_ISDIRECT(dip) && !PM_ISBC(dip)) {
7780 		if (PM_NUMCMPTS(dip) > 2) {
7781 			info->pmi_lp = kmem_alloc(PM_NUMCMPTS(dip) *
7782 			    sizeof (int), KM_SLEEP);
7783 			ip = info->pmi_lp;
7784 		} else {
7785 			ip = info->pmi_levels;
7786 		}
7787 		/* autopm and processes are stopped, ok not to lock power */
7788 		for (i = 0; i < PM_NUMCMPTS(dip); i++)
7789 			*ip++ = PM_CURPOWER(dip, i);
7790 		/*
7791 		 * There is a small window between stopping the
7792 		 * processes and setting pm_processes_stopped where
7793 		 * a driver could get hung up in a pm_raise_power()
7794 		 * call.  Free any such driver now.
7795 		 */
7796 		pm_proceed(dip, PMP_RELEASE, -1, -1);
7797 	}
7798 
7799 	return (DDI_WALK_CONTINUE);
7800 }
7801 
7802 void
7803 pm_restore_direct_levels(void)
7804 {
7805 	/*
7806 	 * If cpr didn't call pm_save_direct_levels, (because stopping user
7807 	 * threads failed) then we don't want to try to restore them
7808 	 */
7809 	if (!pm_processes_stopped)
7810 		return;
7811 
7812 	ddi_walk_devs(ddi_root_node(), pm_restore_direct_lvl_walk, 0);
7813 	pm_processes_stopped = 0;
7814 }
7815 
7816 static int
7817 pm_restore_direct_lvl_walk(dev_info_t *dip, void *arg)
7818 {
7819 	_NOTE(ARGUNUSED(arg))
7820 	PMD_FUNC(pmf, "restore_direct_lvl_walk")
7821 	int i, nc, result;
7822 	int *ip;
7823 
7824 	pm_info_t *info = PM_GET_PM_INFO(dip);
7825 	if (!info)
7826 		return (DDI_WALK_CONTINUE);
7827 
7828 	if (PM_ISDIRECT(dip) && !PM_ISBC(dip)) {
7829 		if ((nc = PM_NUMCMPTS(dip)) > 2) {
7830 			ip = &info->pmi_lp[nc - 1];
7831 		} else {
7832 			ip = &info->pmi_levels[nc - 1];
7833 		}
7834 		/*
7835 		 * Because fb drivers fail attempts to turn off the
7836 		 * fb when the monitor is on, but treat a request to
7837 		 * turn on the monitor as a request to turn on the
7838 		 * fb too, we process components in descending order
7839 		 * Because autopm is disabled and processes aren't
7840 		 * running, it is ok to examine current power outside
7841 		 * of the power lock
7842 		 */
7843 		for (i = nc - 1; i >= 0; i--, ip--) {
7844 			if (PM_CURPOWER(dip, i) == *ip)
7845 				continue;
7846 			if (pm_set_power(dip, i, *ip, PM_LEVEL_EXACT,
7847 			    PM_CANBLOCK_BYPASS, 0, &result) !=
7848 				DDI_SUCCESS) {
7849 				cmn_err(CE_WARN, "cpr: unable "
7850 				    "to restore power level of "
7851 				    "component %d of directly "
7852 				    "power manged device %s@%s"
7853 				    " to %d",
7854 				    i, PM_NAME(dip),
7855 				    PM_ADDR(dip), *ip);
7856 				PMD(PMD_FAIL, ("%s: failed to restore "
7857 				    "%s@%s(%s#%d)[%d] exact(%d)->%d, "
7858 				    "errno %d\n", pmf, PM_DEVICE(dip), i,
7859 				    PM_CURPOWER(dip, i), *ip, result))
7860 			}
7861 		}
7862 		if (nc > 2) {
7863 			kmem_free(info->pmi_lp, nc * sizeof (int));
7864 			info->pmi_lp = NULL;
7865 		}
7866 	}
7867 	return (DDI_WALK_CONTINUE);
7868 }
7869 
7870 /*
7871  * Stolen from the bootdev module
7872  * attempt to convert a path to a major number
7873  */
7874 static major_t
7875 i_path_to_major(char *path, char *leaf_name)
7876 {
7877 	extern major_t path_to_major(char *pathname);
7878 	major_t maj;
7879 
7880 	if ((maj = path_to_major(path)) == (major_t)-1) {
7881 		maj = ddi_name_to_major(leaf_name);
7882 	}
7883 
7884 	return (maj);
7885 }
7886 
7887 /*
7888  * When user calls rem_drv, we need to forget no-involuntary-power-cycles state
7889  * An entry in the list means that the device is detached, so we need to
7890  * adjust its ancestors as if they had just seen this attach, and any detached
7891  * ancestors need to have their list entries adjusted.
7892  */
7893 void
7894 pm_driver_removed(major_t major)
7895 {
7896 	static void i_pm_driver_removed(major_t major);
7897 
7898 	/*
7899 	 * Serialize removal of drivers. This is to keep ancestors of
7900 	 * a node that is being deleted from getting deleted and added back
7901 	 * with different counters.
7902 	 */
7903 	mutex_enter(&pm_remdrv_lock);
7904 	i_pm_driver_removed(major);
7905 	mutex_exit(&pm_remdrv_lock);
7906 }
7907 
7908 /*
7909  * This routine is called recursively by pm_noinvol_process_ancestors()
7910  */
7911 static void
7912 i_pm_driver_removed(major_t major)
7913 {
7914 	PMD_FUNC(pmf, "driver_removed")
7915 	static void adjust_ancestors(char *, int);
7916 	static int pm_is_noinvol_ancestor(pm_noinvol_t *);
7917 	static void pm_noinvol_process_ancestors(char *);
7918 	pm_noinvol_t *ip, *pp = NULL;
7919 	int wasvolpmd;
7920 	ASSERT(major != (major_t)-1);
7921 	PMD(PMD_NOINVOL, ("%s: %s\n", pmf, ddi_major_to_name(major)))
7922 again:
7923 	rw_enter(&pm_noinvol_rwlock, RW_WRITER);
7924 	for (ip = pm_noinvol_head; ip; pp = ip, ip = ip->ni_next) {
7925 		if (major != ip->ni_major)
7926 			continue;
7927 		/*
7928 		 * If it is an ancestor of no-invol node, which is
7929 		 * not removed, skip it. This is to cover the case of
7930 		 * ancestor removed without removing its descendants.
7931 		 */
7932 		if (pm_is_noinvol_ancestor(ip)) {
7933 			ip->ni_flags |= PMC_DRIVER_REMOVED;
7934 			continue;
7935 		}
7936 		wasvolpmd = ip->ni_wasvolpmd;
7937 		/*
7938 		 * remove the entry from the list
7939 		 */
7940 		if (pp) {
7941 			PMD(PMD_NOINVOL, ("%s: freeing %s, prev is %s\n",
7942 			    pmf, ip->ni_path, pp->ni_path))
7943 			pp->ni_next = ip->ni_next;
7944 		} else {
7945 			PMD(PMD_NOINVOL, ("%s: free %s head\n", pmf,
7946 			    ip->ni_path))
7947 			ASSERT(pm_noinvol_head == ip);
7948 			pm_noinvol_head = ip->ni_next;
7949 		}
7950 		rw_exit(&pm_noinvol_rwlock);
7951 		adjust_ancestors(ip->ni_path, wasvolpmd);
7952 		/*
7953 		 * Had an ancestor been removed before this node, it would have
7954 		 * been skipped. Adjust the no-invol counters for such skipped
7955 		 * ancestors.
7956 		 */
7957 		pm_noinvol_process_ancestors(ip->ni_path);
7958 		kmem_free(ip->ni_path, ip->ni_size);
7959 		kmem_free(ip, sizeof (*ip));
7960 		goto again;
7961 	}
7962 	rw_exit(&pm_noinvol_rwlock);
7963 }
7964 
7965 /*
7966  * returns 1, if *aip is a ancestor of a no-invol node
7967  *	   0, otherwise
7968  */
7969 static int
7970 pm_is_noinvol_ancestor(pm_noinvol_t *aip)
7971 {
7972 	pm_noinvol_t *ip;
7973 
7974 	ASSERT(strlen(aip->ni_path) != 0);
7975 	for (ip = pm_noinvol_head; ip; ip = ip->ni_next) {
7976 		if (ip == aip)
7977 			continue;
7978 		/*
7979 		 * To be an ancestor, the path must be an initial substring of
7980 		 * the descendent, and end just before a '/' in the
7981 		 * descendent's path.
7982 		 */
7983 		if ((strstr(ip->ni_path, aip->ni_path) == ip->ni_path) &&
7984 		    (ip->ni_path[strlen(aip->ni_path)] == '/'))
7985 			return (1);
7986 	}
7987 	return (0);
7988 }
7989 
7990 #define	PM_MAJOR(dip) ddi_name_to_major(ddi_binding_name(dip))
7991 /*
7992  * scan through the pm_noinvolpm list adjusting ancestors of the current
7993  * node;  Modifies string *path.
7994  */
7995 static void
7996 adjust_ancestors(char *path, int wasvolpmd)
7997 {
7998 	PMD_FUNC(pmf, "adjust_ancestors")
7999 	char *cp;
8000 	pm_noinvol_t *lp;
8001 	pm_noinvol_t *pp = NULL;
8002 	major_t locked = (major_t)UINT_MAX;
8003 	dev_info_t *dip;
8004 	char	*pathbuf;
8005 	size_t pathbuflen = strlen(path) + 1;
8006 
8007 	/*
8008 	 * First we look up the ancestor's dip.  If we find it, then we
8009 	 * adjust counts up the tree
8010 	 */
8011 	PMD(PMD_NOINVOL, ("%s: %s wasvolpmd %d\n", pmf, path, wasvolpmd))
8012 	pathbuf = kmem_alloc(pathbuflen, KM_SLEEP);
8013 	(void) strcpy(pathbuf, path);
8014 	cp = strrchr(pathbuf, '/');
8015 	if (cp == NULL)	{
8016 		/* if no ancestors, then nothing to do */
8017 		kmem_free(pathbuf, pathbuflen);
8018 		return;
8019 	}
8020 	*cp = '\0';
8021 	dip = pm_name_to_dip(pathbuf, 1);
8022 	if (dip != NULL) {
8023 		locked = PM_MAJOR(dip);
8024 
8025 		(void) pm_noinvol_update(PM_BP_NOINVOL_REMDRV, 0, wasvolpmd,
8026 		    path, dip);
8027 
8028 		if (locked != (major_t)UINT_MAX)
8029 			ddi_release_devi(dip);
8030 	} else {
8031 		char *apath;
8032 		size_t len = strlen(pathbuf) + 1;
8033 		int  lock_held = 1;
8034 
8035 		/*
8036 		 * Now check for ancestors that exist only in the list
8037 		 */
8038 		apath = kmem_alloc(len, KM_SLEEP);
8039 		(void) strcpy(apath, pathbuf);
8040 		rw_enter(&pm_noinvol_rwlock, RW_WRITER);
8041 		for (lp = pm_noinvol_head; lp; pp = lp, lp = lp->ni_next) {
8042 			/*
8043 			 * This can only happen once.  Since we have to drop
8044 			 * the lock, we need to extract the relevant info.
8045 			 */
8046 			if (strcmp(pathbuf, lp->ni_path) == 0) {
8047 				PMD(PMD_NOINVOL, ("%s: %s no %d -> %d\n", pmf,
8048 				    lp->ni_path, lp->ni_noinvolpm,
8049 				    lp->ni_noinvolpm - 1))
8050 				lp->ni_noinvolpm--;
8051 				if (wasvolpmd && lp->ni_volpmd) {
8052 					PMD(PMD_NOINVOL, ("%s: %s vol %d -> "
8053 					    "%d\n", pmf, lp->ni_path,
8054 					    lp->ni_volpmd, lp->ni_volpmd - 1))
8055 					lp->ni_volpmd--;
8056 				}
8057 				/*
8058 				 * remove the entry from the list, if there
8059 				 * are no more no-invol descendants and node
8060 				 * itself is not a no-invol node.
8061 				 */
8062 				if (!(lp->ni_noinvolpm ||
8063 				    (lp->ni_flags & PMC_NO_INVOL))) {
8064 					ASSERT(lp->ni_volpmd == 0);
8065 					if (pp) {
8066 						PMD(PMD_NOINVOL, ("%s: freeing "
8067 						    "%s, prev is %s\n", pmf,
8068 						    lp->ni_path, pp->ni_path))
8069 						pp->ni_next = lp->ni_next;
8070 					} else {
8071 						PMD(PMD_NOINVOL, ("%s: free %s "
8072 						    "head\n", pmf, lp->ni_path))
8073 						ASSERT(pm_noinvol_head == lp);
8074 						pm_noinvol_head = lp->ni_next;
8075 					}
8076 					lock_held = 0;
8077 					rw_exit(&pm_noinvol_rwlock);
8078 					adjust_ancestors(apath, wasvolpmd);
8079 					/* restore apath */
8080 					(void) strcpy(apath, pathbuf);
8081 					kmem_free(lp->ni_path, lp->ni_size);
8082 					kmem_free(lp, sizeof (*lp));
8083 				}
8084 				break;
8085 			}
8086 		}
8087 		if (lock_held)
8088 			rw_exit(&pm_noinvol_rwlock);
8089 		adjust_ancestors(apath, wasvolpmd);
8090 		kmem_free(apath, len);
8091 	}
8092 	kmem_free(pathbuf, pathbuflen);
8093 }
8094 
8095 /*
8096  * Do no-invol processing for any ancestors i.e. adjust counters of ancestors,
8097  * which were skipped even though their drivers were removed.
8098  */
8099 static void
8100 pm_noinvol_process_ancestors(char *path)
8101 {
8102 	pm_noinvol_t *lp;
8103 
8104 	rw_enter(&pm_noinvol_rwlock, RW_READER);
8105 	for (lp = pm_noinvol_head; lp; lp = lp->ni_next) {
8106 		if (strstr(path, lp->ni_path) &&
8107 		    (lp->ni_flags & PMC_DRIVER_REMOVED)) {
8108 			rw_exit(&pm_noinvol_rwlock);
8109 			i_pm_driver_removed(lp->ni_major);
8110 			return;
8111 		}
8112 	}
8113 	rw_exit(&pm_noinvol_rwlock);
8114 }
8115 
8116 /*
8117  * Returns true if (detached) device needs to be kept up because it exported the
8118  * "no-involuntary-power-cycles" property or we're pretending it did (console
8119  * fb case) or it is an ancestor of such a device and has used up the "one
8120  * free cycle" allowed when all such leaf nodes have voluntarily powered down
8121  * upon detach.  In any event, we need an exact hit on the path or we return
8122  * false.
8123  */
8124 int
8125 pm_noinvol_detached(char *path)
8126 {
8127 	PMD_FUNC(pmf, "noinvol_detached")
8128 	pm_noinvol_t *ip;
8129 	int ret = 0;
8130 
8131 	rw_enter(&pm_noinvol_rwlock, RW_READER);
8132 	for (ip = pm_noinvol_head; ip; ip = ip->ni_next) {
8133 		if (strcmp(path, ip->ni_path) == 0) {
8134 			if (ip->ni_flags & PMC_CONSOLE_FB) {
8135 				PMD(PMD_NOINVOL | PMD_CFB, ("%s: inhibits CFB "
8136 				    "%s\n", pmf, path))
8137 				ret = 1;
8138 				break;
8139 			}
8140 #ifdef	DEBUG
8141 			if (ip->ni_noinvolpm != ip->ni_volpmd)
8142 				PMD(PMD_NOINVOL, ("%s: (%d != %d) inhibits %s"
8143 				    "\n", pmf, ip->ni_noinvolpm, ip->ni_volpmd,
8144 				    path))
8145 #endif
8146 			ret = (ip->ni_noinvolpm != ip->ni_volpmd);
8147 			break;
8148 		}
8149 	}
8150 	rw_exit(&pm_noinvol_rwlock);
8151 	return (ret);
8152 }
8153 
8154 int
8155 pm_is_cfb(dev_info_t *dip)
8156 {
8157 	return (dip == cfb_dip);
8158 }
8159 
8160 #ifdef	DEBUG
8161 /*
8162  * Return true if all components of the console frame buffer are at
8163  * "normal" power, i.e., fully on.  For the case where the console is not
8164  * a framebuffer, we also return true
8165  */
8166 int
8167 pm_cfb_is_up(void)
8168 {
8169 	return (pm_cfb_comps_off == 0);
8170 }
8171 #endif
8172 
8173 /*
8174  * Preventing scan from powering down the node by incrementing the
8175  * kidsupcnt.
8176  */
8177 void
8178 pm_hold_power(dev_info_t *dip)
8179 {
8180 	e_pm_hold_rele_power(dip, 1);
8181 }
8182 
8183 /*
8184  * Releasing the hold by decrementing the kidsupcnt allowing scan
8185  * to power down the node if all conditions are met.
8186  */
8187 void
8188 pm_rele_power(dev_info_t *dip)
8189 {
8190 	e_pm_hold_rele_power(dip, -1);
8191 }
8192 
8193 /*
8194  * A wrapper of pm_all_to_normal() to power up a dip
8195  * to its normal level
8196  */
8197 int
8198 pm_powerup(dev_info_t *dip)
8199 {
8200 	PMD_FUNC(pmf, "pm_powerup")
8201 
8202 	PMD(PMD_ALLNORM, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
8203 	ASSERT(!(servicing_interrupt()));
8204 
8205 	/*
8206 	 * in case this node is not already participating pm
8207 	 */
8208 	if (!PM_GET_PM_INFO(dip)) {
8209 		if (!DEVI_IS_ATTACHING(dip))
8210 			return (DDI_SUCCESS);
8211 		if (pm_start(dip) != DDI_SUCCESS)
8212 			return (DDI_FAILURE);
8213 		if (!PM_GET_PM_INFO(dip))
8214 			return (DDI_SUCCESS);
8215 	}
8216 
8217 	return (pm_all_to_normal(dip, PM_CANBLOCK_BLOCK));
8218 }
8219 
8220 int
8221 pm_rescan_walk(dev_info_t *dip, void *arg)
8222 {
8223 	_NOTE(ARGUNUSED(arg))
8224 
8225 	if (!PM_GET_PM_INFO(dip) || PM_ISBC(dip))
8226 		return (DDI_WALK_CONTINUE);
8227 
8228 	/*
8229 	 * Currently pm_cpr_callb/resume code is the only caller
8230 	 * and it needs to make sure that stopped scan get
8231 	 * reactivated. Otherwise, rescan walk needn't reactive
8232 	 * stopped scan.
8233 	 */
8234 	pm_scan_init(dip);
8235 
8236 	(void) pm_rescan(dip);
8237 	return (DDI_WALK_CONTINUE);
8238 }
8239 
8240 static dev_info_t *
8241 pm_get_next_descendent(dev_info_t *dip, dev_info_t *tdip)
8242 {
8243 	dev_info_t *wdip, *pdip;
8244 
8245 	for (wdip = tdip; wdip != dip; wdip = pdip) {
8246 		pdip = ddi_get_parent(wdip);
8247 		if (pdip == dip)
8248 			return (wdip);
8249 	}
8250 	return (NULL);
8251 }
8252 
8253 int
8254 pm_busop_bus_power(dev_info_t *dip, void *impl_arg, pm_bus_power_op_t op,
8255     void *arg, void *result)
8256 {
8257 	PMD_FUNC(pmf, "bp_bus_power")
8258 	dev_info_t	*cdip;
8259 	pm_info_t	*cinfo;
8260 	pm_bp_child_pwrchg_t	*bpc;
8261 	pm_sp_misc_t		*pspm;
8262 	pm_bp_nexus_pwrup_t *bpn;
8263 	pm_bp_child_pwrchg_t new_bpc;
8264 	pm_bp_noinvol_t *bpi;
8265 	dev_info_t *tdip;
8266 	char *pathbuf;
8267 	int		ret = DDI_SUCCESS;
8268 	int		errno = 0;
8269 	pm_component_t *cp;
8270 
8271 	PMD(PMD_SET, ("%s: %s@%s(%s#%d) %s\n", pmf, PM_DEVICE(dip),
8272 	    pm_decode_op(op)))
8273 	switch (op) {
8274 	case BUS_POWER_CHILD_PWRCHG:
8275 		bpc = (pm_bp_child_pwrchg_t *)arg;
8276 		pspm = (pm_sp_misc_t *)bpc->bpc_private;
8277 		tdip = bpc->bpc_dip;
8278 		cdip = pm_get_next_descendent(dip, tdip);
8279 		cinfo = PM_GET_PM_INFO(cdip);
8280 		if (cdip != tdip) {
8281 			/*
8282 			 * If the node is an involved parent, it needs to
8283 			 * power up the node as it is needed.  There is nothing
8284 			 * else the framework can do here.
8285 			 */
8286 			if (PM_WANTS_NOTIFICATION(cdip)) {
8287 				PMD(PMD_SET, ("%s: call bus_power for "
8288 				    "%s@%s(%s#%d)\n", pmf, PM_DEVICE(cdip)))
8289 				return ((*PM_BUS_POWER_FUNC(cdip))(cdip,
8290 				    impl_arg, op, arg, result));
8291 			}
8292 			ASSERT(pspm->pspm_direction == PM_LEVEL_UPONLY ||
8293 			    pspm->pspm_direction == PM_LEVEL_DOWNONLY ||
8294 			    pspm->pspm_direction == PM_LEVEL_EXACT);
8295 			/*
8296 			 * we presume that the parent needs to be up in
8297 			 * order for the child to change state (either
8298 			 * because it must already be on if the child is on
8299 			 * (and the pm_all_to_normal_nexus() will be a nop)
8300 			 * or because it will need to be on for the child
8301 			 * to come on; so we make the call regardless
8302 			 */
8303 			pm_hold_power(cdip);
8304 			if (cinfo) {
8305 				pm_canblock_t canblock = pspm->pspm_canblock;
8306 				ret = pm_all_to_normal_nexus(cdip, canblock);
8307 				if (ret != DDI_SUCCESS) {
8308 					pm_rele_power(cdip);
8309 					return (ret);
8310 				}
8311 			}
8312 			PMD(PMD_SET, ("%s: walk down to %s@%s(%s#%d)\n", pmf,
8313 			    PM_DEVICE(cdip)))
8314 			ret = pm_busop_bus_power(cdip, impl_arg, op, arg,
8315 			    result);
8316 			pm_rele_power(cdip);
8317 		} else {
8318 			ret = pm_busop_set_power(cdip, impl_arg, op, arg,
8319 			    result);
8320 		}
8321 		return (ret);
8322 
8323 	case BUS_POWER_NEXUS_PWRUP:
8324 		bpn = (pm_bp_nexus_pwrup_t *)arg;
8325 		pspm = (pm_sp_misc_t *)bpn->bpn_private;
8326 
8327 		if (!e_pm_valid_info(dip, NULL) ||
8328 		    !e_pm_valid_comp(dip, bpn->bpn_comp, &cp) ||
8329 		    !e_pm_valid_power(dip, bpn->bpn_comp, bpn->bpn_level)) {
8330 			PMD(PMD_SET, ("%s: %s@%s(%s#%d) has no pm info; EIO\n",
8331 			    pmf, PM_DEVICE(dip)))
8332 			*pspm->pspm_errnop = EIO;
8333 			*(int *)result = DDI_FAILURE;
8334 			return (DDI_FAILURE);
8335 		}
8336 
8337 		ASSERT(bpn->bpn_dip == dip);
8338 		PMD(PMD_SET, ("%s: nexus powerup for %s@%s(%s#%d)\n", pmf,
8339 		    PM_DEVICE(dip)))
8340 		new_bpc.bpc_dip = dip;
8341 		pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
8342 		new_bpc.bpc_path = ddi_pathname(dip, pathbuf);
8343 		new_bpc.bpc_comp = bpn->bpn_comp;
8344 		new_bpc.bpc_olevel = PM_CURPOWER(dip, bpn->bpn_comp);
8345 		new_bpc.bpc_nlevel = bpn->bpn_level;
8346 		new_bpc.bpc_private = bpn->bpn_private;
8347 		((pm_sp_misc_t *)(new_bpc.bpc_private))->pspm_direction =
8348 		    PM_LEVEL_UPONLY;
8349 		((pm_sp_misc_t *)(new_bpc.bpc_private))->pspm_errnop =
8350 		    &errno;
8351 		ret = pm_busop_set_power(dip, impl_arg, BUS_POWER_CHILD_PWRCHG,
8352 		    (void *)&new_bpc, result);
8353 		kmem_free(pathbuf, MAXPATHLEN);
8354 		return (ret);
8355 
8356 	case BUS_POWER_NOINVOL:
8357 		bpi = (pm_bp_noinvol_t *)arg;
8358 		tdip = bpi->bpni_dip;
8359 		cdip = pm_get_next_descendent(dip, tdip);
8360 
8361 		/* In case of rem_drv, the leaf node has been removed */
8362 		if (cdip == NULL)
8363 			return (DDI_SUCCESS);
8364 
8365 		cinfo = PM_GET_PM_INFO(cdip);
8366 		if (cdip != tdip) {
8367 			if (PM_WANTS_NOTIFICATION(cdip)) {
8368 				PMD(PMD_NOINVOL,
8369 				    ("%s: call bus_power for %s@%s(%s#%d)\n",
8370 				    pmf, PM_DEVICE(cdip)))
8371 				ret = (*PM_BUS_POWER_FUNC(cdip))
8372 				    (cdip, NULL, op, arg, result);
8373 				if ((cinfo) && (ret == DDI_SUCCESS))
8374 					(void) pm_noinvol_update_node(cdip,
8375 					    bpi);
8376 				return (ret);
8377 			} else {
8378 				PMD(PMD_NOINVOL,
8379 				    ("%s: walk down to %s@%s(%s#%d)\n", pmf,
8380 				    PM_DEVICE(cdip)))
8381 				ret = pm_busop_bus_power(cdip, NULL, op,
8382 				    arg, result);
8383 				/*
8384 				 * Update the current node.
8385 				 */
8386 				if ((cinfo) && (ret == DDI_SUCCESS))
8387 					(void) pm_noinvol_update_node(cdip,
8388 					    bpi);
8389 				return (ret);
8390 			}
8391 		} else {
8392 			/*
8393 			 * For attach, detach, power up:
8394 			 * Do nothing for leaf node since its
8395 			 * counts are already updated.
8396 			 * For CFB and driver removal, since the
8397 			 * path and the target dip passed in is up to and incl.
8398 			 * the immediate ancestor, need to do the update.
8399 			 */
8400 			PMD(PMD_NOINVOL, ("%s: target %s@%s(%s#%d) is "
8401 			    "reached\n", pmf, PM_DEVICE(cdip)))
8402 			if (cinfo && ((bpi->bpni_cmd == PM_BP_NOINVOL_REMDRV) ||
8403 			    (bpi->bpni_cmd == PM_BP_NOINVOL_CFB)))
8404 				(void) pm_noinvol_update_node(cdip, bpi);
8405 			return (DDI_SUCCESS);
8406 		}
8407 
8408 	default:
8409 		PMD(PMD_SET, ("%s: operation %d is not supported!\n", pmf, op))
8410 		return (DDI_FAILURE);
8411 	}
8412 }
8413 
8414 static int
8415 pm_busop_set_power(dev_info_t *dip, void *impl_arg, pm_bus_power_op_t op,
8416     void *arg, void *resultp)
8417 {
8418 	_NOTE(ARGUNUSED(impl_arg))
8419 	PMD_FUNC(pmf, "bp_set_power")
8420 	pm_ppm_devlist_t *devl;
8421 	int clevel, circ;
8422 #ifdef	DEBUG
8423 	int circ_db, ccirc_db;
8424 #endif
8425 	int ret = DDI_SUCCESS;
8426 	dev_info_t *cdip;
8427 	pm_bp_child_pwrchg_t *bpc = (pm_bp_child_pwrchg_t *)arg;
8428 	pm_sp_misc_t *pspm = (pm_sp_misc_t *)bpc->bpc_private;
8429 	pm_canblock_t canblock = pspm->pspm_canblock;
8430 	int scan = pspm->pspm_scan;
8431 	int comp = bpc->bpc_comp;
8432 	int olevel = bpc->bpc_olevel;
8433 	int nlevel = bpc->bpc_nlevel;
8434 	int comps_off_incr = 0;
8435 	dev_info_t *pdip = ddi_get_parent(dip);
8436 	int dodeps;
8437 	int direction = pspm->pspm_direction;
8438 	int *errnop = pspm->pspm_errnop;
8439 	char *dir = pm_decode_direction(direction);
8440 	int *iresp = (int *)resultp;
8441 	time_t	idletime, thresh;
8442 	pm_component_t *cp = PM_CP(dip, comp);
8443 	int work_type;
8444 
8445 	*iresp = DDI_SUCCESS;
8446 	*errnop = 0;
8447 	ASSERT(op == BUS_POWER_CHILD_PWRCHG);
8448 	PMD(PMD_SET, ("%s: %s@%s(%s#%d) %s\n", pmf, PM_DEVICE(dip),
8449 	    pm_decode_op(op)))
8450 
8451 	/*
8452 	 * The following set of conditions indicate we are here to handle a
8453 	 * driver's pm_[raise|lower]_power request, but the device is being
8454 	 * power managed (PM_DIRECT_PM) by a user process.  For that case
8455 	 * we want to pm_block and pass a status back to the caller based
8456 	 * on whether the controlling process's next activity on the device
8457 	 * matches the current request or not.  This distinction tells
8458 	 * downstream functions to avoid calling into a driver or changing
8459 	 * the framework's power state.  To actually block, we need:
8460 	 *
8461 	 * PM_ISDIRECT(dip)
8462 	 *	no reason to block unless a process is directly controlling dev
8463 	 * direction != PM_LEVEL_EXACT
8464 	 *	EXACT is used by controlling proc's PM_SET_CURRENT_POWER ioctl
8465 	 * !pm_processes_stopped
8466 	 *	don't block if controlling proc already be stopped for cpr
8467 	 * canblock != PM_CANBLOCK_BYPASS
8468 	 *	our caller must not have explicitly prevented blocking
8469 	 */
8470 	if (direction != PM_LEVEL_EXACT && canblock != PM_CANBLOCK_BYPASS) {
8471 		PM_LOCK_DIP(dip);
8472 		while (PM_ISDIRECT(dip) && !pm_processes_stopped) {
8473 			/* releases dip lock */
8474 			ret = pm_busop_match_request(dip, bpc);
8475 			if (ret == EAGAIN) {
8476 				PM_LOCK_DIP(dip);
8477 				continue;
8478 			}
8479 			return (*iresp = ret);
8480 		}
8481 		PM_UNLOCK_DIP(dip);
8482 	}
8483 	/* BC device is never scanned, so power will stick until we are done */
8484 	if (PM_ISBC(dip) && comp != 0 && nlevel != 0 &&
8485 	    direction != PM_LEVEL_DOWNONLY) {
8486 		int nrmpwr0 = pm_get_normal_power(dip, 0);
8487 		if (pm_set_power(dip, 0, nrmpwr0, direction,
8488 		    canblock, 0, resultp) != DDI_SUCCESS) {
8489 			/* *resultp set by pm_set_power */
8490 			return (DDI_FAILURE);
8491 		}
8492 	}
8493 	if (PM_WANTS_NOTIFICATION(pdip)) {
8494 		PMD(PMD_SET, ("%s: pre_notify %s@%s(%s#%d) for child "
8495 		    "%s@%s(%s#%d)\n", pmf, PM_DEVICE(pdip), PM_DEVICE(dip)))
8496 		ret = (*PM_BUS_POWER_FUNC(pdip))(pdip, NULL,
8497 		    BUS_POWER_PRE_NOTIFICATION, bpc, resultp);
8498 		if (ret != DDI_SUCCESS) {
8499 			PMD(PMD_SET, ("%s: failed to pre_notify %s@%s(%s#%d)\n",
8500 			    pmf, PM_DEVICE(pdip)))
8501 			return (DDI_FAILURE);
8502 		}
8503 	} else {
8504 		/*
8505 		 * Since we don't know what the actual power level is,
8506 		 * we place a power hold on the parent no matter what
8507 		 * component and level is changing.
8508 		 */
8509 		pm_hold_power(pdip);
8510 	}
8511 	PM_LOCK_POWER(dip, &circ);
8512 	clevel = PM_CURPOWER(dip, comp);
8513 	PMD(PMD_SET, ("%s: %s@%s(%s#%d), cmp=%d, olvl=%d, nlvl=%d, clvl=%d, "
8514 	    "dir=%s\n", pmf, PM_DEVICE(dip), comp, bpc->bpc_olevel, nlevel,
8515 	    clevel, dir))
8516 	switch (direction) {
8517 	case PM_LEVEL_UPONLY:
8518 		/* Powering up */
8519 		if (clevel >= nlevel) {
8520 			PMD(PMD_SET, ("%s: current level is already "
8521 			    "at or above the requested level.\n", pmf))
8522 			*iresp = DDI_SUCCESS;
8523 			ret = DDI_SUCCESS;
8524 			goto post_notify;
8525 		}
8526 		break;
8527 	case PM_LEVEL_EXACT:
8528 		/* specific level request */
8529 		if (clevel == nlevel && !PM_ISBC(dip)) {
8530 			PMD(PMD_SET, ("%s: current level is already "
8531 			    "at the requested level.\n", pmf))
8532 			*iresp = DDI_SUCCESS;
8533 			ret = DDI_SUCCESS;
8534 			goto post_notify;
8535 		} else if (PM_IS_CFB(dip) && (nlevel < clevel)) {
8536 			PMD(PMD_CFB, ("%s: powerdown of console\n", pmf))
8537 			if (!pm_cfb_enabled) {
8538 				PMD(PMD_ERROR | PMD_CFB,
8539 				    ("%s: !pm_cfb_enabled, fails\n", pmf))
8540 				*errnop = EINVAL;
8541 				*iresp = DDI_FAILURE;
8542 				ret = DDI_FAILURE;
8543 				goto post_notify;
8544 			}
8545 			mutex_enter(&pm_cfb_lock);
8546 			while (cfb_inuse) {
8547 				mutex_exit(&pm_cfb_lock);
8548 				if (delay_sig(1) == EINTR) {
8549 					ret = DDI_FAILURE;
8550 					*iresp = DDI_FAILURE;
8551 					*errnop = EINTR;
8552 					goto post_notify;
8553 				}
8554 				mutex_enter(&pm_cfb_lock);
8555 			}
8556 			mutex_exit(&pm_cfb_lock);
8557 		}
8558 		break;
8559 	case PM_LEVEL_DOWNONLY:
8560 		/* Powering down */
8561 		thresh = cur_threshold(dip, comp);
8562 		idletime = gethrestime_sec() - cp->pmc_timestamp;
8563 		if (scan && ((PM_KUC(dip) != 0) ||
8564 		    (cp->pmc_busycount > 0) ||
8565 		    ((idletime < thresh) && !PM_IS_PID(dip)))) {
8566 #ifdef	DEBUG
8567 			if (DEVI(dip)->devi_pm_kidsupcnt != 0)
8568 				PMD(PMD_SET, ("%s: scan failed: "
8569 				    "kidsupcnt != 0\n", pmf))
8570 			if (cp->pmc_busycount > 0)
8571 				PMD(PMD_SET, ("%s: scan failed: "
8572 				    "device become busy\n", pmf))
8573 			if (idletime < thresh)
8574 				PMD(PMD_SET, ("%s: scan failed: device "
8575 				    "hasn't been idle long enough\n", pmf))
8576 #endif
8577 			*iresp = DDI_FAILURE;
8578 			*errnop = EBUSY;
8579 			ret = DDI_FAILURE;
8580 			goto post_notify;
8581 		} else if (clevel != PM_LEVEL_UNKNOWN && clevel <= nlevel) {
8582 			PMD(PMD_SET, ("%s: current level is already at "
8583 			    "or below the requested level.\n", pmf))
8584 			*iresp = DDI_SUCCESS;
8585 			ret = DDI_SUCCESS;
8586 			goto post_notify;
8587 		}
8588 		break;
8589 	}
8590 
8591 	if (PM_IS_CFB(dip) && (comps_off_incr =
8592 	    calc_cfb_comps_incr(dip, comp, clevel, nlevel)) > 0) {
8593 		/*
8594 		 * Pre-adjust pm_cfb_comps_off if lowering a console fb
8595 		 * component from full power.  Remember that we tried to
8596 		 * lower power in case it fails and we need to back out
8597 		 * the adjustment.
8598 		 */
8599 		update_comps_off(comps_off_incr, dip);
8600 		PMD(PMD_CFB, ("%s: %s@%s(%s#%d)[%d] %d->%d cfb_comps_off->%d\n",
8601 		    pmf, PM_DEVICE(dip), comp, clevel, nlevel,
8602 		    pm_cfb_comps_off))
8603 	}
8604 
8605 	if ((*iresp = power_dev(dip,
8606 	    comp, nlevel, clevel, canblock, &devl)) == DDI_SUCCESS) {
8607 #ifdef DEBUG
8608 		/*
8609 		 * All descendents of this node should already be powered off.
8610 		 */
8611 		if (PM_CURPOWER(dip, comp) == 0) {
8612 			pm_desc_pwrchk_t pdpchk;
8613 			pdpchk.pdpc_dip = dip;
8614 			pdpchk.pdpc_par_involved = PM_WANTS_NOTIFICATION(dip);
8615 			ndi_devi_enter(dip, &circ_db);
8616 			for (cdip = ddi_get_child(dip); cdip != NULL;
8617 			    cdip = ddi_get_next_sibling(cdip)) {
8618 				ndi_devi_enter(cdip, &ccirc_db);
8619 				ddi_walk_devs(cdip, pm_desc_pwrchk_walk,
8620 				    (void *)&pdpchk);
8621 				ndi_devi_exit(cdip, ccirc_db);
8622 			}
8623 			ndi_devi_exit(dip, circ_db);
8624 		}
8625 #endif
8626 		/*
8627 		 * Post-adjust pm_cfb_comps_off if we brought an fb component
8628 		 * back up to full power.
8629 		 */
8630 		if (PM_IS_CFB(dip) && comps_off_incr < 0) {
8631 			update_comps_off(comps_off_incr, dip);
8632 			PMD(PMD_CFB, ("%s: %s@%s(%s#%d)[%d] %d->%d "
8633 			    "cfb_comps_off->%d\n", pmf, PM_DEVICE(dip),
8634 			    comp, clevel, nlevel, pm_cfb_comps_off))
8635 		}
8636 		dodeps = 0;
8637 		if (POWERING_OFF(clevel, nlevel)) {
8638 			if (PM_ISBC(dip)) {
8639 				dodeps = (comp == 0);
8640 			} else {
8641 				int i;
8642 				dodeps = 1;
8643 				for (i = 0; i < PM_NUMCMPTS(dip); i++) {
8644 					/* if some component still on */
8645 					if (PM_CURPOWER(dip, i)) {
8646 						dodeps = 0;
8647 						break;
8648 					}
8649 				}
8650 			}
8651 			if (dodeps)
8652 				work_type = PM_DEP_WK_POWER_OFF;
8653 		} else if (POWERING_ON(clevel, nlevel)) {
8654 			if (PM_ISBC(dip)) {
8655 				dodeps = (comp == 0);
8656 			} else {
8657 				int i;
8658 				dodeps = 1;
8659 				for (i = 0; i < PM_NUMCMPTS(dip); i++) {
8660 					if (i == comp)
8661 						continue;
8662 					if (PM_CURPOWER(dip, i) > 0) {
8663 						dodeps = 0;
8664 						break;
8665 					}
8666 				}
8667 			}
8668 			if (dodeps)
8669 				work_type = PM_DEP_WK_POWER_ON;
8670 		}
8671 
8672 		if (dodeps) {
8673 			char *pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
8674 
8675 			(void) ddi_pathname(dip, pathbuf);
8676 			pm_dispatch_to_dep_thread(work_type, pathbuf, NULL,
8677 			    PM_DEP_NOWAIT, NULL, 0);
8678 			kmem_free(pathbuf, MAXPATHLEN);
8679 		}
8680 		if ((PM_CURPOWER(dip, comp) == nlevel) && pm_watchers()) {
8681 			int old;
8682 
8683 			/* If old power cached during deadlock, use it. */
8684 			old = (cp->pmc_flags & PM_PHC_WHILE_SET_POWER ?
8685 			    cp->pmc_phc_pwr : olevel);
8686 			mutex_enter(&pm_rsvp_lock);
8687 			pm_enqueue_notify(PSC_HAS_CHANGED, dip, comp, nlevel,
8688 			    old, canblock);
8689 			pm_enqueue_notify_others(&devl, canblock);
8690 			mutex_exit(&pm_rsvp_lock);
8691 		}
8692 
8693 		/*
8694 		 * If we are coming from a scan, don't do it again,
8695 		 * else we can have infinite loops.
8696 		 */
8697 		if (!scan)
8698 			pm_rescan(dip);
8699 	} else {
8700 		/* if we incremented pm_comps_off_count, but failed */
8701 		if (comps_off_incr > 0) {
8702 			update_comps_off(-comps_off_incr, dip);
8703 			PMD(PMD_CFB, ("%s: %s@%s(%s#%d)[%d] %d->%d "
8704 			    "cfb_comps_off->%d\n", pmf, PM_DEVICE(dip),
8705 			    comp, clevel, nlevel, pm_cfb_comps_off))
8706 		}
8707 		*errnop = EIO;
8708 	}
8709 
8710 post_notify:
8711 	/*
8712 	 * This thread may have been in deadlock with pm_power_has_changed.
8713 	 * Before releasing power lock, clear the flag which marks this
8714 	 * condition.
8715 	 */
8716 	cp->pmc_flags &= ~PM_PHC_WHILE_SET_POWER;
8717 
8718 	/*
8719 	 * Update the old power level in the bus power structure with the
8720 	 * actual power level before the transition was made to the new level.
8721 	 * Some involved parents depend on this information to keep track of
8722 	 * their children's power transition.
8723 	 */
8724 	if (*iresp != DDI_FAILURE)
8725 		bpc->bpc_olevel = clevel;
8726 
8727 	if (PM_WANTS_NOTIFICATION(pdip)) {
8728 		ret = (*PM_BUS_POWER_FUNC(pdip))(pdip, NULL,
8729 		    BUS_POWER_POST_NOTIFICATION, bpc, resultp);
8730 		PM_UNLOCK_POWER(dip, circ);
8731 		PMD(PMD_SET, ("%s: post_notify %s@%s(%s#%d) for "
8732 		    "child %s@%s(%s#%d), ret=%d\n", pmf, PM_DEVICE(pdip),
8733 		    PM_DEVICE(dip), ret))
8734 	} else {
8735 		nlevel = cur_power(cp); /* in case phc deadlock updated pwr */
8736 		PM_UNLOCK_POWER(dip, circ);
8737 		/*
8738 		 * Now that we know what power transition has occurred
8739 		 * (if any), release the power hold.  Leave the hold
8740 		 * in effect in the case of OFF->ON transition.
8741 		 */
8742 		if (!(clevel == 0 && nlevel > 0 &&
8743 		    (!PM_ISBC(dip) || comp == 0)))
8744 			pm_rele_power(pdip);
8745 		/*
8746 		 * If the power transition was an ON->OFF transition,
8747 		 * remove the power hold from the parent.
8748 		 */
8749 		if ((clevel > 0 || clevel == PM_LEVEL_UNKNOWN) &&
8750 		    nlevel == 0 && (!PM_ISBC(dip) || comp == 0))
8751 			pm_rele_power(pdip);
8752 	}
8753 	if (*iresp != DDI_SUCCESS || ret != DDI_SUCCESS)
8754 		return (DDI_FAILURE);
8755 	else
8756 		return (DDI_SUCCESS);
8757 }
8758 
8759 /*
8760  * If an app (SunVTS or Xsun) has taken control, then block until it
8761  * gives it up or makes the requested power level change, unless
8762  * we have other instructions about blocking.  Returns DDI_SUCCESS,
8763  * DDI_FAILURE or EAGAIN (owner released device from directpm).
8764  */
8765 static int
8766 pm_busop_match_request(dev_info_t *dip, void *arg)
8767 {
8768 	PMD_FUNC(pmf, "bp_match_request")
8769 	pm_bp_child_pwrchg_t *bpc = (pm_bp_child_pwrchg_t *)arg;
8770 	pm_sp_misc_t *pspm = (pm_sp_misc_t *)bpc->bpc_private;
8771 	int comp = bpc->bpc_comp;
8772 	int nlevel = bpc->bpc_nlevel;
8773 	pm_canblock_t canblock = pspm->pspm_canblock;
8774 	int direction = pspm->pspm_direction;
8775 	int clevel, circ;
8776 
8777 	ASSERT(PM_IAM_LOCKING_DIP(dip));
8778 	PM_LOCK_POWER(dip, &circ);
8779 	clevel = PM_CURPOWER(dip, comp);
8780 	PMD(PMD_SET, ("%s: %s@%s(%s#%d), cmp=%d, nlvl=%d, clvl=%d\n",
8781 	    pmf, PM_DEVICE(dip), comp, nlevel, clevel))
8782 	if (direction == PM_LEVEL_UPONLY) {
8783 		if (clevel >= nlevel) {
8784 			PM_UNLOCK_POWER(dip, circ);
8785 			PM_UNLOCK_DIP(dip);
8786 			return (DDI_SUCCESS);
8787 		}
8788 	} else if (clevel == nlevel) {
8789 		PM_UNLOCK_POWER(dip, circ);
8790 		PM_UNLOCK_DIP(dip);
8791 		return (DDI_SUCCESS);
8792 	}
8793 	if (canblock == PM_CANBLOCK_FAIL) {
8794 		PM_UNLOCK_POWER(dip, circ);
8795 		PM_UNLOCK_DIP(dip);
8796 		return (DDI_FAILURE);
8797 	}
8798 	if (canblock == PM_CANBLOCK_BLOCK) {
8799 		/*
8800 		 * To avoid a deadlock, we must not hold the
8801 		 * power lock when we pm_block.
8802 		 */
8803 		PM_UNLOCK_POWER(dip, circ);
8804 		PMD(PMD_SET, ("%s: blocking\n", pmf))
8805 		    /* pm_block releases dip lock */
8806 		    switch (pm_block(dip, comp, nlevel, clevel)) {
8807 		    case PMP_RELEASE:
8808 				return (EAGAIN);
8809 		    case PMP_SUCCEED:
8810 				return (DDI_SUCCESS);
8811 		    case PMP_FAIL:
8812 				return (DDI_FAILURE);
8813 		    }
8814 	} else {
8815 		ASSERT(0);
8816 	}
8817 	_NOTE(NOTREACHED);
8818 	return (DDI_FAILURE);	/* keep gcc happy */
8819 }
8820 
8821 static int
8822 pm_all_to_normal_nexus(dev_info_t *dip, pm_canblock_t canblock)
8823 {
8824 	PMD_FUNC(pmf, "all_to_normal_nexus")
8825 	int		*normal;
8826 	int		i, ncomps;
8827 	size_t		size;
8828 	int		changefailed = 0;
8829 	int		ret, result = DDI_SUCCESS;
8830 	pm_bp_nexus_pwrup_t	bpn;
8831 	pm_sp_misc_t	pspm;
8832 
8833 	ASSERT(PM_GET_PM_INFO(dip));
8834 	PMD(PMD_ALLNORM, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
8835 	if (pm_get_norm_pwrs(dip, &normal, &size) != DDI_SUCCESS) {
8836 		PMD(PMD_ALLNORM, ("%s: can't get norm pwrs\n", pmf))
8837 		return (DDI_FAILURE);
8838 	}
8839 	ncomps = PM_NUMCMPTS(dip);
8840 	for (i = 0; i < ncomps; i++) {
8841 		bpn.bpn_dip = dip;
8842 		bpn.bpn_comp = i;
8843 		bpn.bpn_level = normal[i];
8844 		pspm.pspm_canblock = canblock;
8845 		pspm.pspm_scan = 0;
8846 		bpn.bpn_private = &pspm;
8847 		ret = pm_busop_bus_power(dip, NULL, BUS_POWER_NEXUS_PWRUP,
8848 		    (void *)&bpn, (void *)&result);
8849 		if (ret != DDI_SUCCESS || result != DDI_SUCCESS) {
8850 			PMD(PMD_FAIL | PMD_ALLNORM, ("%s: %s@%s(%s#%d)[%d] "
8851 			    "->%d failure result %d\n", pmf, PM_DEVICE(dip),
8852 			    i, normal[i], result))
8853 			changefailed++;
8854 		}
8855 	}
8856 	kmem_free(normal, size);
8857 	if (changefailed) {
8858 		PMD(PMD_FAIL, ("%s: failed to set %d comps %s@%s(%s#%d) "
8859 		    "full power\n", pmf, changefailed, PM_DEVICE(dip)))
8860 		return (DDI_FAILURE);
8861 	}
8862 	return (DDI_SUCCESS);
8863 }
8864 
8865 int
8866 pm_noinvol_update(int subcmd, int volpmd, int wasvolpmd, char *path,
8867     dev_info_t *tdip)
8868 {
8869 	PMD_FUNC(pmf, "noinvol_update")
8870 	pm_bp_noinvol_t args;
8871 	int ret;
8872 	int result = DDI_SUCCESS;
8873 
8874 	args.bpni_path = path;
8875 	args.bpni_dip = tdip;
8876 	args.bpni_cmd = subcmd;
8877 	args.bpni_wasvolpmd = wasvolpmd;
8878 	args.bpni_volpmd = volpmd;
8879 	PMD(PMD_NOINVOL, ("%s: update for path %s tdip %p subcmd %d "
8880 	    "volpmd %d wasvolpmd %d\n", pmf,
8881 	    path, (void *)tdip, subcmd, wasvolpmd, volpmd))
8882 	ret = pm_busop_bus_power(ddi_root_node(), NULL, BUS_POWER_NOINVOL,
8883 	    &args, &result);
8884 	return (ret);
8885 }
8886 
8887 void
8888 pm_noinvol_update_node(dev_info_t *dip, pm_bp_noinvol_t *req)
8889 {
8890 	PMD_FUNC(pmf, "noinvol_update_node")
8891 
8892 	PMD(PMD_NOINVOL, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
8893 	switch (req->bpni_cmd) {
8894 	case PM_BP_NOINVOL_ATTACH:
8895 		PMD(PMD_NOINVOL, ("%s: PM_PB_NOINVOL_ATTACH %s@%s(%s#%d) "
8896 		    "noinvol %d->%d\n", pmf, PM_DEVICE(dip),
8897 		    DEVI(dip)->devi_pm_noinvolpm,
8898 		    DEVI(dip)->devi_pm_noinvolpm - 1))
8899 		ASSERT(DEVI(dip)->devi_pm_noinvolpm);
8900 		PM_LOCK_DIP(dip);
8901 		DEVI(dip)->devi_pm_noinvolpm--;
8902 		if (req->bpni_wasvolpmd) {
8903 			PMD(PMD_NOINVOL, ("%s: PM_BP_NOINVOL_ATTACH "
8904 			    "%s@%s(%s#%d) volpmd %d->%d\n", pmf,
8905 			    PM_DEVICE(dip), DEVI(dip)->devi_pm_volpmd,
8906 			    DEVI(dip)->devi_pm_volpmd - 1))
8907 			if (DEVI(dip)->devi_pm_volpmd)
8908 				DEVI(dip)->devi_pm_volpmd--;
8909 		}
8910 		PM_UNLOCK_DIP(dip);
8911 		break;
8912 
8913 	case PM_BP_NOINVOL_DETACH:
8914 		PMD(PMD_NOINVOL, ("%s: PM_BP_NOINVOL_DETACH %s@%s(%s#%d) "
8915 		    "noinvolpm %d->%d\n", pmf, PM_DEVICE(dip),
8916 		    DEVI(dip)->devi_pm_noinvolpm,
8917 		    DEVI(dip)->devi_pm_noinvolpm + 1))
8918 		PM_LOCK_DIP(dip);
8919 		DEVI(dip)->devi_pm_noinvolpm++;
8920 		if (req->bpni_wasvolpmd) {
8921 			PMD(PMD_NOINVOL, ("%s: PM_BP_NOINVOL_DETACH "
8922 			    "%s@%s(%s#%d) volpmd %d->%d\n", pmf,
8923 			    PM_DEVICE(dip), DEVI(dip)->devi_pm_volpmd,
8924 			    DEVI(dip)->devi_pm_volpmd + 1))
8925 			DEVI(dip)->devi_pm_volpmd++;
8926 		}
8927 		PM_UNLOCK_DIP(dip);
8928 		break;
8929 
8930 	case PM_BP_NOINVOL_REMDRV:
8931 		PMD(PMD_NOINVOL, ("%s: PM_BP_NOINVOL_REMDRV %s@%s(%s#%d) "
8932 		    "noinvol %d->%d\n", pmf, PM_DEVICE(dip),
8933 		    DEVI(dip)->devi_pm_noinvolpm,
8934 		    DEVI(dip)->devi_pm_noinvolpm - 1))
8935 		ASSERT(DEVI(dip)->devi_pm_noinvolpm);
8936 		PM_LOCK_DIP(dip);
8937 		DEVI(dip)->devi_pm_noinvolpm--;
8938 		if (req->bpni_wasvolpmd) {
8939 			PMD(PMD_NOINVOL,
8940 			    ("%s: PM_BP_NOINVOL_REMDRV %s@%s(%s#%d) "
8941 			    "volpmd %d->%d\n", pmf, PM_DEVICE(dip),
8942 			    DEVI(dip)->devi_pm_volpmd,
8943 			    DEVI(dip)->devi_pm_volpmd - 1))
8944 			/*
8945 			 * A power up could come in between and
8946 			 * clear the volpmd, if that's the case,
8947 			 * volpmd would be clear.
8948 			 */
8949 			if (DEVI(dip)->devi_pm_volpmd)
8950 				DEVI(dip)->devi_pm_volpmd--;
8951 		}
8952 		PM_UNLOCK_DIP(dip);
8953 		break;
8954 
8955 	case PM_BP_NOINVOL_CFB:
8956 		PMD(PMD_NOINVOL,
8957 		    ("%s: PM_BP_NOIVOL_CFB %s@%s(%s#%d) noinvol %d->%d\n",
8958 		    pmf, PM_DEVICE(dip), DEVI(dip)->devi_pm_noinvolpm,
8959 		    DEVI(dip)->devi_pm_noinvolpm + 1))
8960 		PM_LOCK_DIP(dip);
8961 		DEVI(dip)->devi_pm_noinvolpm++;
8962 		PM_UNLOCK_DIP(dip);
8963 		break;
8964 
8965 	case PM_BP_NOINVOL_POWER:
8966 		PMD(PMD_NOINVOL,
8967 		    ("%s: PM_BP_NOIVOL_PWR %s@%s(%s#%d) volpmd %d->%d\n",
8968 		    pmf, PM_DEVICE(dip),
8969 		    DEVI(dip)->devi_pm_volpmd, DEVI(dip)->devi_pm_volpmd -
8970 		    req->bpni_volpmd))
8971 		PM_LOCK_DIP(dip);
8972 		DEVI(dip)->devi_pm_volpmd -= req->bpni_volpmd;
8973 		PM_UNLOCK_DIP(dip);
8974 		break;
8975 
8976 	default:
8977 		break;
8978 	}
8979 
8980 }
8981 
8982 #ifdef DEBUG
8983 static int
8984 pm_desc_pwrchk_walk(dev_info_t *dip, void *arg)
8985 {
8986 	PMD_FUNC(pmf, "desc_pwrchk")
8987 	pm_desc_pwrchk_t *pdpchk = (pm_desc_pwrchk_t *)arg;
8988 	pm_info_t *info = PM_GET_PM_INFO(dip);
8989 	int i, curpwr, ce_level;
8990 
8991 	if (!info)
8992 		return (DDI_WALK_CONTINUE);
8993 
8994 	PMD(PMD_SET, ("%s: %s@%s(%s#%d)\n", pmf, PM_DEVICE(dip)))
8995 	for (i = 0; i < PM_NUMCMPTS(dip); i++) {
8996 		if ((curpwr = PM_CURPOWER(dip, i)) == 0)
8997 			continue;
8998 		ce_level = (pdpchk->pdpc_par_involved == 0) ? CE_PANIC :
8999 		    CE_WARN;
9000 		PMD(PMD_SET, ("%s: %s@%s(%s#%d) is powered off while desc "
9001 		    "%s@%s(%s#%d)[%d] is at %d\n", pmf,
9002 		    PM_DEVICE(pdpchk->pdpc_dip), PM_DEVICE(dip), i, curpwr))
9003 		cmn_err(ce_level, "!device %s@%s(%s#%d) is powered on, "
9004 		    "while its ancestor, %s@%s(%s#%d), is powering off!",
9005 		    PM_DEVICE(dip), PM_DEVICE(pdpchk->pdpc_dip));
9006 	}
9007 	return (DDI_WALK_CONTINUE);
9008 }
9009 #endif
9010 
9011 /*
9012  * Record the fact that one thread is borrowing the lock on a device node.
9013  * Use is restricted to the case where the lending thread will block until
9014  * the borrowing thread (always curthread) completes.
9015  */
9016 void
9017 pm_borrow_lock(kthread_t *lender)
9018 {
9019 	lock_loan_t *prev = &lock_loan_head;
9020 	lock_loan_t *cur = (lock_loan_t *)kmem_zalloc(sizeof (*cur), KM_SLEEP);
9021 
9022 	cur->pmlk_borrower = curthread;
9023 	cur->pmlk_lender = lender;
9024 	mutex_enter(&pm_loan_lock);
9025 	cur->pmlk_next = prev->pmlk_next;
9026 	prev->pmlk_next = cur;
9027 	mutex_exit(&pm_loan_lock);
9028 }
9029 
9030 /*
9031  * Return the borrowed lock.  A thread can borrow only one.
9032  */
9033 void
9034 pm_return_lock(void)
9035 {
9036 	lock_loan_t *cur;
9037 	lock_loan_t *prev = &lock_loan_head;
9038 
9039 	mutex_enter(&pm_loan_lock);
9040 	ASSERT(prev->pmlk_next != NULL);
9041 	for (cur = prev->pmlk_next; cur; prev = cur, cur = cur->pmlk_next)
9042 		if (cur->pmlk_borrower == curthread)
9043 			break;
9044 
9045 	ASSERT(cur != NULL);
9046 	prev->pmlk_next = cur->pmlk_next;
9047 	mutex_exit(&pm_loan_lock);
9048 	kmem_free(cur, sizeof (*cur));
9049 }
9050